YouTube字幕・文字起こしスクレーパー

https://www.youtube.com/watch?v=Ri-HcFlNcJk

iOS 17 Hands on - Top 10 Features!

Mrwhosetheboss

https://www.youtube.com/@Mrwhosetheboss

12-Jul-23

Intro 0:00 this is IOS
 17. I'm using it for about a 0:03 month it's probably the single biggest 0:04
 updates that iPhones have ever had you 0:06 can do multiple timers at once
 you can 0:08 automatically clear out any used 0:09 Verification codes from
 your emails you 0:11 can reply to messages by swiping right 0:13 you can show
 it a photo of a meal and 0:15 it'll tell you how to make it it even 0:16 has
 an AI that can learn to speak in 0:19 your voice which we'll test later and
 0:21 these are just the little things so here 0:22 are the 10 major changes
 that you need 0:24 to know bearing in mind that this is the 0:26 beta
 starting with the basics iOS 17 has 0:28 an upgraded language model or in
 other TEXT CORRECTION 0:30 words your phone will better understand 0:32 what
 you're trying to say and this works 0:34 in two ways autocorrect is
 noticeably 0:36 more accurate like you can actually test 0:38 this side by
 side with an iOS 16 phone 0:40 and see how it just gets those really 0:42
 subtle human nuances and dictation I 0:46 have not tested a phone that can do
 0:47 dictation better than this iOS 17 phone 0:50 can when you talk you can
 see it right 0:51 there like one word behind what you're 0:53 saying just
 waiting to hear your 0:54 intonation so it knows whether to add a 0:57 comma
 or a question mark and even sorts 0:59 the apostrophes play plus swear words
 it 1:01 no longer just assumes that you meant to 1:03 say ducking number nine
 is the FaceTime 1:06 upgrade so when you now react to things FACETIME 1:08
 you'll get these augmented reality 1:10 effects that I guess are just there
 to 1:12 amplify that expression it's not useful 1:14 and I can totally
 imagine you wanting to 1:16 turn the feature off but I do like that 1:18 you
 have to be quite purposeful if you 1:19 want to trigger them they won't
 happen 1:21 automatically that the effects will work 1:22 even though the
 other person might not 1:24 have IOS 17 and that they're actually 1:26 using
 this true depth camera system on 1:27 the front of your phone to figure out
 1:29 where you are in the frame and place the 1:31 effects not just on top of
 you like a 1:32 sticker but around you plus now that 1:34 they've got this
 augmented reality stuff 1:36 going on in FaceTime it also means that 1:38 you
 can do portrait mode effects just 1:40 like you can in your camera app except
 I 1:42 don't use it there and I do use it here 1:44 being able to increase
 the amount of 1:46 light on your face and not just blur out 1:48 the
 background behind you but dark in it 1:50 is like the best way to take a
 video 1:53 call and then because the iPhone's doing 1:54 all this processing
 on the hardware 1:56 level not the software level it works 1:58 across other
 apps too like Zoom and 2:00 WebEx plus you can now leave a FaceTime 2:02
 message if the person doesn't pick up it 2:05 really feels like apple wants
 FaceTime 2:06 to be the way that you call people okay 2:09 personalized
 contact posters is probably CONTACT POSTERS 2:11 the headline feature of iOS
 17. Apple 2:13 spent quite a bit of time at their event 2:15 talking about it
 and it didn't really 2:16 register to me as something that would 2:17 be any
 kind of game changer but it's 2:19 only using it that I'm realizing how 2:21
 smart it is so you pick a name and then 2:23 either a memoji uh me Milo gee a
 photo 2:27 or a letter and then you can fiddle with 2:29 those elements till
 you get to a poster 2:30 that you're happy with now the main 2:32 downside is
 that it's not unlimited 2:34 customization you could definitely do 2:35 more
 with this concept but I imagine the 2:37 reason behind controlling it is to
 2:39 create some sort of consistency so 2:41 everyone's posters follow the
 same 2:43 format so they're recognizable and so 2:45 those same details can
 be used in 2:46 multiple different parts of the UI and I 2:48 really rate
 this feature it feels very 2:50 easy to get a result that looks 2:52
 professional you flip between different 2:53 presets and even if your image
 doesn't 2:55 fill the screen they fade it out in a 2:57 way that makes it
 look purposeful and 2:58 probably the best thing about contact 3:00 posters
 is that it's you picking how you 3:02 come across to everyone else up until
 3:04 this point the best way to make all your 3:06 contacts look neat and
 consistent has 3:07 been you setting the photos and details 3:09 for other
 people I've tried to do this 3:11 one by one a few times on some of my 3:12
 past phones because I do I love the idea 3:15 of a fully organized clean
 contacts list 3:17 but it just takes a special kind of 3:20 commitment to
 actually keep that up 3:21 consistently whereas now each person is 3:24 only
 responsible for one person's image 3:25 and that's you it's how you are going
 to 3:28 look to other people so it's applying 3:30 that little bit of social
 pressure that 3:31 I think you need for a feature like this 3:33 to actually
 kick off it also happens to 3:35 be a very clever way to make iPhone 3:37
 users pressure their Android friends to 3:39 also get iPhones oh that even
 Milo's 3:43 climbed up he doesn't look that good 3:45 normally 3:47 now the
 contact posters also tie in 3:50 really neatly with the new airdrop so 3:52
 what you used to have to do is to open 3:53 the media you wanted to share
 Click 3:55 Share and then click airdrop and 3:56 potentially also who you
 wanted to 3:58 airdrop to now you just bring two iOS 17 4:01 plus phones
 together and the transfer 4:03 initiates it's using NFC to check for 4:05
 other phones which means that it's not 4:06 like wireless charging where you
 have to 4:07 perfectly align two things to an exact 4:10 spot and the way it
 animates is so sick 4:12 first time I discovered it with drisha 4:14 we just
 sat there for 10 minutes doing 4:15 it again and again so we could keep 4:17
 seeing it in action but also they have 4:19 fixed what I would say is the
 main 4:20 problem with airdrop which is that it's 4:22 only so far worked
 when you're close 4:24 essentially when you bring two iPhones 4:25 together
 they find each other via 4:27 Bluetooth and then create a direct Fast 4:29
 Five gigahertz Wi-Fi network between 4:31 them so the phone sending the file
 is 4:33 like a Wi-Fi Hub that the phone 4:34 receiving the file is connecting
 to 4:36 that's why it's so fast that's why you 4:38 don't need to be
 connected to a normal 4:40 Wi-Fi for its work but it's also why if 4:42 you
 step more than about 10 feet away 4:44 from each other it cancels there's
 only 4:45 so far that your small phone can Propel 4:47 that direct Wi-Fi
 signal so what happens 4:50 in iOS 17 is that as soon as you pull 4:52 your
 devices far enough away that the 4:54 direct phone to phone connection
 Fizzles 4:55 are odds both devices realize and they 4:57 switch their
 connection over to an 4:59 indirect transfer the device sending the 5:01 file
 is just uploading to the internet 5:02 at the same time as the receiving file
 5:04 is downloading from the Internet it's 5:06 slower but slow is better
 than ever but 5:08 then it's not just files you can also 5:10 share your
 contact poster like this 5:12 along with all the other details about 5:14 you
 that you want to so there's now a 5:15 very tangible benefit to each person
 5:17 filling out their own contact details 5:19 and making their poster look
 nice and 5:20 pretty 5:21 now okay there is a Siri upgrade too and SIRI 5:25
 I'm particularly glad that Siri is 5:27 getting some love because it feels
 like 5:28 it was introduced 12 years ago as the 5:30 future of how to
 interact with your 5:31 phone and then it just sat there while 5:34 Google
 Assistant has been getting better 5:35 at a much faster rate so Siri now 5:37
 responds to just the word Siri you don't 5:40 need to say hey anymore 5:42 me
 realizing that I've just accidentally 5:44 triggered every iOS 17 user's
 phone but 5:47 it is quite clever because it waits for 5:49 a split second
 after you finish the word 5:51 to make sure that you're not about to 5:52 say
 cereal or serious we've now got 5:56 continued conversation which to be fair
 5:58 Google Assistant has had for a long time 5:59 but nonetheless I would
 say is the 6:01 single biggest Improvement that Siri has 6:03 had from the
 very beginning because it 6:04 means you can actually have a 6:06
 conversation with it without needing to 6:07 tap the button every single time
 like 6:09 this what's the weather today rain is 6:12 okay what about tomorrow
 looks like 6:15 what about this time next week 6:21 I can't so you just told
 me what it is 6:23 next week 6:23 okay well the continued conversation 6:25
 part of it is cool plus you can ask it 6:27 to read web pages by just saying
 Siri 6:29 read this 6:32 tmau is an uncommon condition that 6:34 causes an
 unpleasant fishy smell 6:37 Siri call Doctor it's not quite like 6:40 real
 speech like it still has that 6:41 robotic intonation that modern AI 6:43
 programs are actually starting to bypass 6:45 but I'm using this to read out
 news 6:47 articles in the mornings and it's not 6:49 too far off feeling like
 a personalized 6:52 one-person radio station oh yeah and I 6:54 spent 15
 minutes last night rattling off 6:56 the weirdest phrases that the film is
 6:57 asking me to do so that it could train 6:59 to learn my voice a
 gentleman with the 7:01 fan exclaimed good morning 7:04 what is this and to test
 the results in 7:07 action hello there my name is iron Mani 7:10 I'm a 27
 year old economics graduate and 7:12 I love phones what I never said those
 7:16 words to this phone 7:18 number five though has got to be the 7:21
 Safari update so for starters you can SAFARI 7:23 make profiles like work and
 fun it's a 7:25 bit convoluted you actually have to go 7:27 into the settings
 to make those profiles 7:28 before you can use this but once you've 7:30 done
 that you can flick between these 7:31 different modes by tapping this icon I
 7:33 would say this itself is one of those 7:35 hyper specialized features
 that I 7:36 probably won't use because you already 7:38 have tab groups which
 can group all the 7:40 tabs related to any one thing together 7:42 but what
 is cool is that they've also 7:44 made the tab groups experience better 7:46
 too you used to have to switch between 7:47 them with this not so pretty menu
 now 7:49 you just swipe so when you're in a tab 7:51 and you swipe it swipes
 to the next tab 7:53 in the tab group you're in and then when 7:54 you zoom
 out to look at all your tabs in 7:56 the group you can swipe to change the
 7:58 group and the other thing which now I've 8:00 seen it as a feature just
 feels like 8:01 such a no-brainer your private browsing 8:03 windows are
 locked by default so no one 8:05 else can see them unless they have your 8:07
 face which it changes the dynamic from 8:09 making sure that you always close
 every 8:11 single one of those private tabs after 8:13 using them to now
 trusting that it 8:15 doesn't matter only you'll be able to 8:16 see them
 regardless but then how much 8:18 bigger change has got to be this new 8:20
 standby mode so as long as your phone is STANDBY 8:22 locked on charge in
 some way it can be 8:24 Apple's 100 plus wireless charging stand 8:26 which
 is very fancy but I'm glad that it 8:28 doesn't have to be that and that you
 8:30 just make sure it's in horizontal mode 8:31 it'll activate this new
 interface 8:33 there's a widget screen which lets you 8:34 pick from a bunch
 of different 8:35 interactive titles a photos page and 8:37 then a clock page
 where you can cycle 8:39 through different clock Styles it 8:41 actually
 feels a lot like an Apple Watch 8:43 to use now I don't think this is for
 8:45 everyone it's kind of everything your 8:46 phone already does but just
 present it 8:48 in a different way but there is certain 8:50 contexts where I
 do see the benefit like 8:53 if you're working for example and you 8:54 want
 to keep an eye on your phone in 8:55 case something important comes up but
 8:57 you don't want to be on your phone then 8:58 sticking it a bit further
 away from you 9:00 in this standby mode it feels like a 9:02 more passive way
 to keep up to date kind 9:04 of like that nothing phone we just 9:05 tested
 I'll leave that video linked from 9:07 this one I'm liking this new attention
 9:08 from filmmakers towards mindful use of 9:11 the smartphone and probably
 the best 9:13 part of it is that if you're really into 9:14 sports being able
 to see live scores 9:16 without actually having to find a place 9:17 to watch
 it and the distraction element 9:19 of that I think that's great oh and it
 9:22 has automatic night mode you know how 9:23 you get those blue light
 filter apps 9:25 that take out a lot of a distracting 9:26 blue light that
 wakes you up and strains 9:28 your eyes well a night mode here there 9:30 is
 no blue light and hey if you're 9:32 enjoying this video then a sub to the
 9:33 channel would be IO yes 9:37 I don't know 9:39 the interactable widgets
 do not end with WIDGETS 9:42 standby though so this is an iOS 17 home 9:44
 screen you can call someone directly 9:46 from it and I was quite surprised
 to see 9:47 you can configure it so this left hand 9:49 button over here for
 example launches a 9:51 FaceTime video but then the right hand 9:52 button
 launches a WhatsApp message you 9:55 can play and pause music you can control
 9:56 your podcasts it's all pretty simple 9:58 stuff but I'm a big believer
 in widgets 10:01 like these because they keep you out of 10:02 apps and the
 webs of algorithms that 10:05 those apps use to make you lose track of 10:07
 time but by far the thing that I'm most 10:09 excited about in iOS 17 is
 what's 10:11 happening with messages so for starters IMESSAGE 10:13 new
 interface very shiny but then you 10:16 know the speech detections just got
 10:17 better so now when you send a voice note 10:20 it literally instantly
 transcribes it 10:22 and it's smart about it like if you send 10:25 a 15
 minute recap of your life it knows 10:27 that that's something that the other
 10:28 person needs to listen to to get but if 10:30 you just wanted to send a
 voice note 10:31 that says hey remember to buy milk 10:33 because say you're
 in a situation where 10:36 you can't type then it will turn that 10:37
 message into text so the other person 10:39 can get the contents of that
 message in 10:41 whatever the most convenient way is for 10:43 them at that
 time it's a subtle thing 10:45 but I think it matters and then the 10:47
 cherry on top is check-in which is where 10:49 your phone uses its location
 data to let 10:51 the people you care about know 10:52 automatically when
 you've reached where 10:54 you told them you were going which saves 10:56 you
 having to do the whole text me when 10:57 you get there okay I've arrived
 dance 10:59 every single time but to be really 11:01 honest more so than any
 of the features 11:03 that are actually useful I have had the 11:06 most fun
 playing around with stickers I STICKERS 11:08 have not once in my life made a
 custom 11:10 sticker on a phone 11:12 until iOS 17 because this makes it very
 11:14 easy and very very cool so let's say 11:17 you're browsing your photos
 and you come 11:18 across this masterpiece you just hold 11:20 down on the
 face and click create 11:21 sticker that's it that's something that 11:24 you
 can now drop straight away into 11:25 messages and not just in this really
 11:27 flat way that feels like a typical 11:29 conversation thread you can
 put them 11:30 anywhere and then you can turn those 11:32 digital stickers
 into what feels like 11:34 physical stickers with different effects 11:36
 that respond to how you tilt your phone 11:38 I gasped when I saw this not
 because 11:41 it's bleeding edge Tech but just because 11:43 it's a really
 clever human feeling and 11:45 direction that leverages the tech you 11:47
 already have now I will say it does feel 11:49 a little at odds with the very
 polished 11:51 controlled nature of some of the 11:52 iPhone's other features
 like contact 11:54 posters because when you start messing 11:55 with stickers
 these chats get very 11:57 chaotic very quickly but then I'd be 12:00 lying
 if I said it didn't allow you to 12:01 express yourself better than you used
 to 12:03 be able to like if I think about the 12:04 absolute whale of a time
 that my team 12:06 has had making custom emojis for our 12:07 slack group
 this is a playground on a 12:10 whole other tier and that's iOS 17. I'm 12:12
 kind of sad to see that there's nothing 12:14 major new for the dynamic
 Island 12:15 considering that is one of the newest 12:17 Hardware features
 but the overall 12:19 direction I like and I want to keep 12:22 making iOS
 videos like this as well as 12:23 summaries of what's happening in the 12:24
 world of Android so let me know if you 12:27 want to see that too

https://www.youtube.com/watch?v=ej9lpaE3LiI

iOS 17: All NEW Features You Need to Know!

MacRumors

https://www.youtube.com/@macrumors

18-Sep-23

Intro 0:00 iOS 17 is
 officially available for 0:02 everyone you can go into your settings 0:04 app
 you can go under software and it 0:06 should pop right up and it's a big 0:08
 update that's jam-packed with tons of 0:10 new features in this video we're
 going 0:12 to go over some of the features that I 0:14 think you need to know
 about it's not 0:16 all of the features and if you want to 0:17 know more
 features and information you 0:19 can always check the link in the 0:21
 description down below you should have a 0:23 whole list there but we're
 going to try 0:24 to run through some of these pretty 0:25 quickly in order
 to keep this video 0:27 relatively short so let's start off with Phone App
 0:30 the phone app the phone app now gives 0:31 you the ability to customize
 how you 0:33 appear on other people's devices when 0:35 you call them with
 your own custom 0:37 poster and you can make all of the 0:39 tweaks you want
 inside of the phone app 0:41 once you go ahead and start editing your 0:43
 custom poster and there's also a new 0:45 live voicemail feature which is one
 of 0:47 my favorite new features that Apple has 0:48 introduced with iOS 17
 and it gives 0:51 users a live transcription as someone 0:53 starts to leave
 you a voicemail message 0:55 and you can actually read the message as 0:57
 it's happening and then decide whether 0:58 or not you want to still pick up
 the 0:59 phone call or let the person continue 1:02 the message 1:03 in
 FaceTime you can now leave a video or 1:06 audio message to capture exactly
 what 1:08 you want to say to somebody when they 1:09 actually don't pick up
 your FaceTime 1:10 call and you can also make FaceTime 1:13 calls using your
 iPhone on your Apple TV 1:16 so you use your iPhone as a camera you 1:18 get
 a little Mount there and you stick 1:19 it on top of your TV and now you can
 1:21 have a FaceTime call with the whole 1:23 family if you want to while
 using your 1:25 iPhone as the camera standby is a new Standby 1:28 feature that
 turns your iPhone into a 1:31 home hub when docked to a charger and if 1:34
 you turn it horizontally this feature 1:36 offers a full screen experience
 with 1:37 glanceable information like clocks 1:39 photos and widgets designed
 to be viewed 1:42 from a distance in places like your 1:45 nightstand or a
 kitchen counter or your 1:47 desk now there are tons of different 1:49
 widgets and clocks and different things 1:50 that you can add to it you can
 see your 1:52 photo library and I just I really love 1:55 this feature it's
 honestly something 1:56 that I wasn't anticipating with iOS 17 1:59 but it
 has quickly become one of my 2:01 favorites and if you use a Max save 2:03
 charger the feature will actually 2:04 automatically remember your preferred
 2:06 View and it'll just revert back to that 2:08 option whenever you place
 it on a MAG 2:09 safe charger there are finally 2:11 interactive widgets
 available that let 2:13 you take actions like Mark a reminder as 2:16
 complete turn off a light in the home 2:18 app all directly from the widget
 in 2:20 either the home screen lock screen or in 2:22 standby the messages
 app got a ton of Messages 2:25 new features but here are a couple that 2:26
 are worth mentioning live stickers can 2:29 now be created by lifting the
 subject 2:30 from photos and videos and you can turn 2:33 them into stickers
 with stylized effects 2:35 like shiny puffy comic and outline there 2:38 are
 also better search improvements to 2:40 help find messages faster you can
 swipe 2:43 right to reply to a message in line and 2:45 the iMessage apps now
 have this very 2:47 nice new UI that just makes the keyboard 2:49 area far
 more minimal and less cluttered 2:53 speaking of the keyboard many good 2:55
 quality of life improvements here like 2:56 easier auto correct editing which
 2:59 temporarily underlines corrected words 3:01 and lets you revert back to
 what you 3:03 originally typed with just a tap and 3:05 inline predictive
 text shows single and 3:07 multi-word predictions as you type that 3:10 can
 be added by tapping the space bar 3:12 one of my favorite features are the
 3:14 Verification codes that automatically 3:15 pop up when you get messages
 that's like 3:17 one of the most underrated things or 3:19 maybe now it's
 properly rated uh but 3:21 with iOS 17 you can actually get those 3:24 codes
 from emails as well it's not just 3:26 SMS so if a code pops up in your email
 3:28 it'll actually pop up on the keyboard 3:30 like it normally does
 whenever you get 3:32 one of those sent via SMS and you can 3:33 just
 automatically tap it and it'll fill 3:35 it in it's honestly a huge huge
 boost to 3:39 an already great feature in the music Music 3:41 app share play
 makes it easy for 3:43 everyone to control and play Apple music 3:45 in the
 car and Crossfade smoothly 3:48 transitions between songs by fading out 3:50
 the currently playing song while fading 3:52 in the next one so that the
 music just 3:54 never stops Other 3:55 there's a new airdrop feature called
 3:57 name drop which lets you exchange your 4:00 contact information by just
 bringing two 4:02 phones together like this and the 4:04 information will
 automatically be sent 4:05 to the other person's device and you get 4:08 this
 cool little animation that just 4:09 makes it look really awesome and it 4:11
 works super well in the maps app you can 4:14 finally get offline maps which
 allows 4:16 you to select an area you want to access 4:18 search and explore
 Rich information for 4:21 places to download for use when your 4:23 iPhone
 doesn't have Wi-Fi or cellular 4:25 signal and there are also some new 4:27
 airpods Pro 2 features like adaptive 4:29 audio which blends A and C and 4:31
 transparency to tailor the noise control 4:34 experience and along those same
 lines 4:35 you get personalized volume which 4:37 adjusts the volume of your
 media in 4:39 response to your environment and the 4:41 same can be done with
 conversation 4:42 awareness which also tailors the volume 4:44 of your media
 and it enhances voices 4:46 when a conversation is detected and 4:49 again
 these are not all of the iOS 17 4:51 features but these are just the ones
 4:52 that I think are pretty important but 4:54 there are tons of others and
 again you 4:55 can check that link in the description 4:57 down below if you want
 to see all of the 4:59 iOS 17 features but of course I'd love 5:01 to hear
 from you in the comments down 5:02 below what do you think of iOS 17 now 5:04
 that it's officially available what's 5:06 your favorite new feature let me
 know 5:08 down in those comments this has been Dan 5:09 with Mac Rumors
 thanks so much for 5:11 watching and I hope to see you around in 5:12 the
 next video

https://www.youtube.com/watch?v=4Hy__KNNWK8&t=13s

How to Set up Pagination and Loop in Octoparse

Octoparse

https://www.youtube.com/@Octoparsewebscraping

14-Sep-25

0:00 Octopass, an easy
 web scraper for 0:03 anyone. 0:05 Hi everyone, welcome back to Octopass 0:07
 channel. In the previous session, we 0:10 focused on setting up a basic data
 0:11 collection task with pagenation and loop 0:14 playing a particularly
 important role. 0:16 In this session, we'll dive deeper into 0:18 these two
 features, discovering how 0:20 pagenation and loop can provide clever 0:21
 ways to expand your workflows and 0:23 achieve more with your data collection.
 0:25 In our previous exercises, you might 0:27 observe that the tip pane
 frequently 0:29 suggests pageionation methods while 0:31 you're customizing a
 task. Pageionation 0:33 is the process of dividing content into 0:35 separate
 pages commonly seen on websites 0:37 that list items, articles, or search
 0:40 results. Now, let's take a closer look 0:42 at what these three
 pageionation types 0:44 actually are and see how you can set 0:46 them up
 manually in the workflow 0:47 designer if you prefer not to rely on 0:49 the
 tip pane. 0:51 First of all is the next page button 0:53 method. The next
 page method is the 0:55 traditional form of pageionation. It is 0:58 used
 when the website has a clear next 0:59 or arrow button for pageionation. 1:02
 For example, as you can see on the 1:04 screen, this eBay page is a typical
 1:06 example of such a layout where you can 1:08 click that next button to
 move from one 1:10 page to the next to load more product 1:12 listings
 information. In Octopenation 1:15 for this kind of pages pattern website 1:17
 is straightforward. You can simply add a 1:19 loop and place the click action
 inside 1:21 it. Then place the cursor in the right 1:23 place. The input box
 has already 1:25 generated an X path in it. In this way, 1:27 the task will
 automatically perform the 1:29 pageionation step repeatedly navigating 1:32
 through all pages without manual 1:33 intervention. 1:35 Furthermore, the
 number of page turns is 1:37 controlled in the general section under 1:39 the
 loop option. The number of repeats 1:41 determines how many pages will be
 1:43 turned. That's how we set up 1:44 pageionation for this kind of page in
 1:46 the workflow designer. 1:48 If you want to do it in the browser 1:50
 area, you can also click the next page 1:52 button, select the loop, click in
 the 1:55 tips pane. A pageionation loop shows up. 1:59 The second approach is
 the load more 2:01 button method. In this case, 2:03 pageionation requires
 the user to click 2:05 the designated load area. Once clicked, 2:08
 additional results are appended directly 2:10 to the current page instead of
 2:11 triggering a full reload. In octopar, 2:14 handling a load more
 pagenation pattern 2:16 is largely the same as with the former 2:18 setup.
 2:19 If you want to do it in the browser 2:21 area, you can click the load
 more button 2:24 directly here, then choose loop click, 2:26 and you will see
 the loop instantly 2:28 appear. 2:30 Of course, you can also set up manually
 2:32 in the workflow designer. All you need 2:34 to do is add a loop and drag
 a click 2:36 inside it. However, the crucial 2:38 difference is that it
 requires an 2:39 additional X path for the button 2:41 element. Because the
 load more button 2:43 typically appears only after some 2:45 content has loaded
 and its position on 2:47 the page can vary. The cursorbased 2:49 selection is
 unreliable and we need an X 2:51 path for it. Similarly, you can also 2:54
 control the repeating number of page 2:55 turns in the general section.
 Before we 2:58 go further, there's a quick note on what 2:59 is path. XPath
 is a language used to 3:02 navigate and identify elements within an 3:04 XML
 or HTML document. In web scraping, 3:08 it allows tools like Octopse to 3:10
 precisely locate web elements. Even if 3:12 the button moves around on the
 page, as 3:14 long as its structure in the HTML stays 3:16 the same, XPath
 can still find it. If 3:18 you're new to this, it might feel a bit 3:20
 tricky, but we'll keep it brief for now. 3:22 Don't worry, we'll dive deeper
 in the 3:24 following videos, showing you how to 3:26 write an X path and how
 it can help with 3:27 more sophisticated scraping tasks. In 3:30 this
 website, the X path for the load 3:32 more button just looks like this. we
 can 3:34 simply put it in the XP path input box. 3:37 Lastly, let's come to
 the infinite 3:39 scrolling method. It is also noticeable 3:41 that some
 pages don't have any buttons 3:43 at all yet new content keeps appearing 3:45
 as you scroll down which offers a smooth 3:47 and seamless browsing
 experience. In 3:49 Octoping 3:55 number under the scroll setting, which 3:57
 directly controls how many times the 3:59 page will scroll. That's how you
 set up 4:01 pageionation for this kind of page. So 4:03 far in our
 discussion, pagionation in 4:05 Octopse relies on a loop. And loops can 4:08
 do much more than just turn pages. In 4:10 Octtopse, there are six built-in
 loop 4:12 modes in the workflow designer. Let's 4:14 break down each mode
 step by step to see 4:16 how they work in practice. First up, 4:19 let's talk
 about the single element 4:20 loop. In simple terms, this loop keeps 4:23
 performing the same action on a single 4:24 element until a certain condition
 is 4:26 met. A classic use case is pageionation 4:29 which we covered earlier
 repeatedly 4:31 clicking the single element of the next 4:32 page position.
 Instead of moving through 4:35 a list of items, the crawler keeps 4:37
 repeating the same action on one single 4:38 element until the task is done.
 4:42 Next, let's take a look at the fixed 4:43 list loop. In simple terms,
 this loop is 4:46 meant for lists where the number of 4:47 items is already
 set. Each element has a 4:50 predefined x path and octtop processes 4:52 them
 in order exactly as you specify. 4:55 Fixed list is quite similar to a 4:57
 variable list. It locates a list of 4:59 items which is a list of X path
 queries 5:01 with each X path locating a unique 5:03 element on the page. It
 is used when the 5:05 number of elements on the page is 5:06 consistent
 across all pages. 5:09 As you input the selected fixed list X 5:11 path,
 Octopse will correspondingly 5:13 identify them. It highlights all 5:15
 matching items on the page, creates a 5:17 looping container for them. Right
 now, 5:19 you might find the idea a bit confusing 5:21 for now, mainly
 because we haven't 5:23 touched on XPath in this course yet, but 5:25 we'll
 revisit this concept in a later 5:27 lesson with more details unpacked 5:28
 through customized task examples. 5:31 Now, let's move on to the variable
 list 5:33 loop. Unlike the fixed list, this loop 5:35 is designed for lists
 where the number 5:37 of items can vary. Instead of manually 5:39 defining
 each element, Octopse 5:41 identifies the repeating pattern on the 5:43 page
 and creates a loop that adapts to 5:45 however many items are present. 5:47
 Sometimes you see 10 items, other times 5:49 20 depending on the page. With a
 5:51 variable list loop, Octopar can handle 5:53 both scenarios seamlessly
 without extra 5:55 setup. Inside the variable list, Octopar 5:59 also creates
 a general X path that 6:00 matches all the elements in that list. 6:04
 Another powerful option is the list of 6:06 URL loop. Instead of relying on
 elements 6:08 detected on a page, this loop is driven 6:10 by a predefined
 list of web addresses. 6:13 You can click the small button here to 6:14 input
 your URL listings. Octopse will 6:17 open each URL in the list and process
 6:19 them one by one following the same 6:21 extraction workflow. This loop
 is 6:23 perfect when you already have a set of 6:25 target pages to scrape.
 For example, a 6:27 list of product detail pages, news 6:29 articles, or
 company profiles. No matter 6:32 how different the pages look in 6:33
 navigation, as long as the structure 6:35 inside each page is consistent, you
 can 6:37 apply the same data extraction rules 6:39 across all of them. Then
 comes the text 6:41 list. This mode lets you loop through a 6:43 list of text
 values. It's commonly used 6:46 for entering multiple keywords into a 6:47
 search box or testing multiple input 6:49 values. To set it up, hit the
 search bar 6:52 in the browser and add an enter text 6:54 action in a loop.
 Select enter text and 6:57 loop and just type in your keywords in 6:58 the
 provided bar. Then hit the enter key 7:01 when finished entering which tells
 7:02 octopus to automatically press enter 7:04 after typing in each keyword.
 You will 7:07 see that the workflow designer has 7:08 already generated a
 loop action and 7:10 input the text and loop. Lastly, the 7:12 scroll page
 loop is used for pages that 7:14 load new content as you scroll, such as 7:16
 social media feeds, job boards, or 7:19 e-commerce listings. We have 7:20
 demonstrated the application of scroll 7:22 before. You can set how far and
 how 7:24 often it scrolls or stop when no new 7:26 content appears. 7:29
 That's it. That's the six kinds of loops 7:32 you can use in Octopus to
 automate 7:33 repetitive actions, coupled with the 7:35 smart pageionation
 feature to navigate 7:37 through web pages. Together, these tools 7:40 form
 the foundation of powerful 7:41 workflows, letting you handle data 7:43
 collection with far less effort. In the 7:45 next lesson, we'll dive deeper
 into 7:47 XPath, the backbone of precise data 7:49 extraction. Make sure to
 try out these 7:51 techniques yourself and follow along.

https://www.youtube.com/watch?v=nYXVvK-Wmn0

AI Engineer Roadmap ?€? How to Learn AI in 2025

freeCodeCamp.org

https://www.youtube.com/@freecodecamp

6-Feb-25

AI Engineering Roadmap
 Introduction 0:00 this AI engineering road map takes you from core
 fundamentals to Advanced AI 0:06 implementations it covers essential
 mathematics machine learning deep learning and large language models 0:14
 providing you with the exact skills needed to thrive as an AI engineer in
 0:19 2025 whether you're starting fresh or upgrading your skills this road
 map offers a clear path to success with 0:27 hands-on experience and Industry
 relevant insights T from lunar Tech 0:32 developed this course imagine being
 at the Forefront of one of the most transformative fields of 0:39 our time
 where technology meets Innovation and changes the world welcome 0:45 to the
 AI engineering road map of 2025 my name is D Vasan from lunar Tech 0:52 and
 I'm absolutely exciting to be here with you today to dive into this highly
 0:57 requested topic together we will will explore everything that you need
 to know to navigate this exciting world of 1:05 artificial intelligence and
 AI engineering to set yourself up for success in this field in this video we
 1:11 are going to break down the step-by-step road map for becoming a
 worldclass AI 1:17 engineer here is what we are going to cover first we will
 Define what AI engineering is and how it feds into this 1:24 broader
 ecosystem of AI and data science next we will explore the real world 1:31
 applications of AI engineering showcasing its really strong power 1:36
 transformative impact across different Industries then we will dive into the
 1:42 must have versus nice to have skills helping you to understand exactly
 where 1:47 to focus your efforts and your time finally we will go to
 step-by-step 1:53 process so the skill sets that you need to master outlining
 the essential topics 1:58 to help you become a job ready AI engineer this
 session is packed with 2:03 unique insights and practical tips that you won't
 find any URS so stay tuned 2:10 without further Ado let's get 2:25 started so
 let's start with the basics what is AI engineering AI engineering is 2:31
 this practice of Designing building and deploying AI systems that solve real
 2:37 world problems it sits in this intersection of software engineering
 machine learning and data science and 2:45 here is how it fits into this
 broader Tech world and the ecosystem so the data 2:50 scientists often focus
 on analyzing data or predicting something or developing 2:55 models AI
 Engineers take these models and make them work in the real world 3:01
 settings and with much more advanced models they create systems that process
 data make decisions and deliver 3:08 actionable insights for example in the
 healthcare a data scientist might develop a machine learning model to 3:14
 detect the tumors in x-rays an AI engineer brings this to the next level 3:20
 he ensures that the model is integrated into Hospital Systems runs in real
 time 3:25 and works reliably under different conditions also AI Engineers
 they work with much more advanced models like deep 3:32 learning models or
 neural network based models so data science principles system 3:38 design
 optimization machine learning deep learning is what all combines into one
 place which is AI engineering it's 3:45 not just about building models it's
 about making sure that those models actually solve problems and deliver 3:51
 value for the business or this public Enterprise and that's why AI
 engineering 3:57 is such a critical role in today's Tech ecosystem it's where
 this Cutting Edge What is AI Engineering 4:03 research meets the Practical
 industry impactful implementation so bridging 4:08 this gap between the
 research and the actual engineering so um AI engineering isn't 4:16 just
 limited to one field it's changing Industries all over the world let's look
 4:22 actually at some of the examples how AI engineering is making an impact
 first up 4:28 is the healthcare so AI systems are used to analyze medical
 images predict 4:34 patients outcomes and also assist the doctors in the drug
 Discovery or the patient care AI engineers build the 4:42 systems to ensure
 that those are scalable reliable and efficient for real 4:47 world use next
 up is the finance from fraud detection to aloric trading AI processes 4:56
 massive amount of financial data in real time engineers in this field they focus
 5:01 on creating secure efficient and realtime systems that can handle this
 sensitive information real time like FR 5:09 detection in the retail and
 e-commerce in the platforms like Amazon they use AI to personalize
 recommendations optimize 5:16 pricing and manage inventory AI Engineers they
 design algorithms and systems that drive this experiences next 5:24 up is the
 entertainment of course the streaming platforms like Netflix they rely on AI
 for personalized content 5:32 recommendations jna tools like Dolly and
 chatbot chbt are changing now how the 5:38 creators produce content next up
 is the autonomous vehicles so self-driving cars they 5:45 depend on AI for
 navigation object detection and decision making AI AI Engineering
 Applications 5:52 Engineers they are the ones who design this algorithms and
 Hardware integration to make this autonomous Vehicle Systems 5:59 safe and
 reliable so these examples are just few of them and they show how 6:05
 different and impactful AI engineering is so whether you are passionate about
 6:10 health care Finance Tech defense or any other creative industry there is
 a place 6:16 for you in this field and that is actually why the AI
 engineering is so popular this day and it's going to be 6:22 one of the most
 independent Professionals in the next decade there are many Industries and
 companies who 6:28 are currently Hing when it comes to the salaries for AI
 Engineers those are 6:34 highly competitive just 40 ENT roll they start
 around 80 up to 6:41 120k at least for the midlevel engineers this is uh 120k
 to 180k in us and where 6:50 senior roles this can take all the way from 200
 up to 750k in the US dollar so let's now get 6:58 into the actual skill set
 that you must know in order to become an AI engineer 7:03 and here I'm
 talking about becoming a worldclass well-rounded real AI engineer 7:08 not
 just someone who does promp engineering real AI engineer not just 7:14
 someone who does promp engineering and without knowing these different models
 uh just uses them but actually becomes 7:21 someone who will create new
 algorithms who will create their own unicorns or will become an AI and
 without knowing 7:28 these different models uh just uses them but actually
 become someone who will create new algorithms who will create 7:35 their own
 unicorns or will become an AI engineer that works at this uh large Cutting
 Edge companies like open AI 7:43 Tesla meta and many other Cutting Edge
 startups so first up is of course the 7:49 mathematics mathematics is a Fiel
 when it comes to traditional machine learning all the way to the most Cutting
 Edge AI 7:56 that you see nowadays so um when it comes to mathematics there
 are different topics from this field that you must 8:02 know not the entire
 universe of mathematics or the super advanced stuff but really the
 fundamentals and um these 8:10 are selected topics from different uh levels
 so you cannot just say first 8:15 level of University or second level of
 University of that specific study no it's a combination of these different
 8:22 levels from this different fields and studies that you need to combine
 in one place learn it such that you can move on 8:30 on to the next page and
 today I will tell you which are those in a more detail such that you are left
 with a Must-Have Skills for an AI Engineer 8:37 specific topics for you in
 mind to learn mathematics if you decided to do a self-study and become an
 self faced AI 8:45 engineer on your own so first up is the high school
 mathematics in here um you 8:50 can understand doing basic divisions how to
 solve an equation with uh squared 8:57 unknowns so for example a square plus
 something you are able to uh calculate 9:02 the discriminant to find the
 solutions to that equation you know this different um geometric um terms like
 what is sinus Mathematical Foundations 9:11 what is cosine what is tangent
 what is cotangent uh the Pythagorean theorem um 9:18 basically all the topics
 from the high school all the way to the last level 9:25 next up is the uh
 linear algebra of course linear Al ra comes usually from 9:31 the second uh
 year of econometric study or applied mathematical and statistical 9:36
 studies and this field is really important for understanding not just the
 traditional machine learning but also 9:43 the Deep learning which is really
 important and it's a more advanced type of ml that powers today's most
 cutting 9:50 gge applications including the GPT models the Transformers Etc
 so if you 9:56 want to know and understand the cycle of n networks the
 training how it's being 10:01 optimized and how this entire neural networks
 structure works then you must 10:07 understand linear algebra so when it
 comes to linear algebra let me tell you specifically what I mean not the
 entire 10:14 linear algebra but really to understand the norm of a vector
 this understanding 10:19 of vector and matrices the cartisian coordinate
 system that comes from um the 10:25 high school but then here is also very
 relevant to understand where the vector are how you can position the vectors
 in 10:31 the cian coordinate system understand this idea of Norm versus alal
 and 10:36 distance the uh Pythagorean theorem here again the orthogonality um
 you also need to 10:43 understand the vectors and operations so foundations
 of the vector the special vectors unit vectors um and also uh the 10:52 idea
 of dot product the application of the dot product the C squares equation
 10:59 also you need to understand the matrices and the solving of the linear
 systems using this idea of matrices so here you 11:06 need to have the
 foundations of linear systems and matrices you need to uh be 11:12 able to
 add matrices multiply them to compute a DOT product between matrices 11:17 or
 between Matrix and a vector um also understanding of ging reduction the 11:24
 reduced ulum form the row reduced ulum form the no space the c space the rank
 11:31 the full rank this all will be foundation for you to understand how
 11:36 this their networks work um if you truly want to understand um the
 different deep 11:42 learning and AI models you also need to have a good
 basis when it comes to 11:47 linear transformation and matrices so this
 algebraic lows for matrices uh 11:53 including how um it actually works how
 you can uh solve a system with the 11:59 linear equations multiple of them
 using these different Transformations so what is for example the transpose of
 a matrix 12:06 what is the inverse of a metrix and apply these different uh
 rows and the rules from linear algebra uh also what 12:14 is the determinant
 how you can calculate it what are the properties of determinant the transpose
 of matrices I 12:20 believe I just mentioned and then you also need to
 understand some topics from Advanced linear algebra like uh the 12:28
 projections of vectors um the gr Schmid process the infamous process that you
 um 12:35 need to understand uh the metrix factorization really important not
 just 12:40 for the Deep learning but also for the traditional machine
 learning or the things like metrix uh factorization that is used in the 12:47
 recommender systems so uh this part is also very important to understand the
 QR 12:52 de composition ion values igon vectors uh which is really important
 for 12:58 understanding the principal comp quasis and dimensionality
 reduction also the igon de composition which is based on 13:04 igon values
 and igon vectors and understand the singular value the composition or the SVD
 which is really 13:11 important part as part of traditional machine learning
 so um this is what uh 13:16 you need to know when it comes to the linear
 algebra and if you are looking for that 13:23 one place to learn linear
 algebra then uh last year uh we have published an 13:29 entire 26 plus hour
 course that covers all these topics in one place it was 13:35 quite a popular
 course uh and highly demanded one and you can get also a 13:40 certification
 once you completed so check out this course the fundamental s linear algebra
 uh at the lunch. to also 13:49 uh go through all these topics uh follow it
 study it practice it and then get 13:54 also a certification next up when it
 comes to mathematics Beyond um the linear algebra 14:01 and the um High
 School mathematics you also need to understand calculus this one is really
 important as well uh you 14:09 will need to have an understanding what are
 the gradients what are the derivatives how you can calculate 14:14
 derivatives how you can calculate the integrals not just with one n but with
 14:20 two variables basically so double integrals um how you can uh use this
 uh 14:26 derivatives and integrals when comes to optimization this uh concept
 of the 14:32 slope and uh optimization of the models using the gradients
 first order gradient 14:37 and second order gradient in the context of it how
 you can adjust the parameters for better 14:43 accuracy and um just a
 traditional calculus one and some calculus 2 so um 14:51 this is um
 no-brainer when it comes to AI not just for advanced AI but for the 14:58
 traditional machine learning learning for understanding these different
 models you must know calculus next up is the 15:05 game theory not the entire
 universe of Game Theory not all the topics but there are some topics from
 Game Theory which 15:11 usually comes from third year of econometrical or ply
 mathematical studies is something that you must know 15:18 think about NES
 equilibrium or the mean Max strategy or this um um this game 15:25 where um
 competing is actually resulting in worse outcome than 15:31 collaborating so
 uh this idea of NES equilibrium is really important for understanding one of
 the foundational 15:39 generative AI models which is the generative
 adversarial networks so for understanding one of this Genna models 15:46 you
 will need to also have this uh couple of topics from game theory in place all
 right so that's about the 15:54 mathematics um and here I'm also not
 mentioning this foundational geometry 15:59 topics which is usually also
 covered as part of high school so once again the sign cosine the tangent how
 to work with 16:07 with the different um angles the 90?? angle what are these
 different values 16:13 for different angles and this common notation with the
 pi so what the pi represents the radians Etc once you 16:20 comfortable with
 this mathematical topics the next topic that I would suggest you to study is
 the statistics 16:26 statistics is very important when it com comes to
 becoming a well-rounded AI professional to understand the um idea 16:35 of
 predicting the next word but all the way to the very basic machine learning
 16:40 uh having this basics of Statistics will be very helpful to you so here
 is the 16:46 list of topics that I would suggest you to study when it comes
 to statistics so first up of course understanding this 16:52 concept of
 probabilities to know what the probabilities are what is its 16:58 concept uh
 why it is used for this concept of probability distribution 17:03 functions
 the PDFs the cumulative distribution functions or the cdfs and also um to
 understand uh what is this 17:11 idea of sample why we use sample um versus
 population um this idea of having a 17:19 representative sample work with the
 data so understanding for example what are the random variables what is this
 idea 17:26 of experiment uh what are the probabilities um the uh criteria and
 17:32 qualities of probabilities what is the PDF or the probability
 distribution function uh what is the cumulative uh Statistics Essentials
 17:39 distribution function this uh basic statistics like the mean the median
 the 17:45 variance the standard deviation the mode um and also how they can
 be calculated 17:51 this um idea of covariance and correlation what is the
 difference between correlation and 17:57 cation uh understanding um how these
 different statistics can be used to describe your 18:04 data and to tell a
 story about your data and um also this idea of Sample versus 18:12 population
 why we use sample um and why we um are unable for example to deal 18:19 with
 a population um and how this becomes relevant when it comes to this entire
 18:26 universe of data science um also understanding the bias theorem the
 18:31 different rules when it comes to the probabilities like the conditional
 probability the idea of Independence 18:38 between different random variables
 um then I get into some Bic probability 18:44 distribution functions especially
 the normal distribution function the baroli distribution function this idea
 of boli 18:51 Trials the binomial distribution function what is this
 connection between bomal distribution function and the 18:56 binomial
 distribution function how it is used in these different concepts like tossing
 a coin so basic statistics 19:04 basically uh also understand uh the idea of
 uh linear regression and ordinary Le 19:11 squares what are these different
 uh conditions and assumptions that this 19:16 ordinary squares is making when
 calculating and optimizing these different um parameter estimates this 19:24
 idea of estimation versus um the unknown parameter the idea of error terms
 the 19:31 error terms versus residuals um and also this concept of gas Mark
 of theorem how it is used um 19:38 and this comes usually from econometrics
 and the idea of parameters what are the properties of parameters like the
 bias 19:45 of a parameter the consistency and the efficiency and this is
 again tied back to the gas Mark of theorem uh also the 19:54 understanding of
 confidence interal will be really important in your career in the field of
 science and AI the idea of 20:01 95% confidence interval how it's calculated
 what is this idea of um 20:06 calculating this interval the lower bound and
 the upper bound what it means another very important topic from 20:13
 statistics is this idea of hypothesis testing why we need hypothesis testing
 the idea of null um uh hypothesis the 20:21 alternative hypothesis how you
 set up these experiments why it is important why we even need it the concept
 of 20:28 statistic iCal significance is very important how to calculate type
 one 20:33 error type two error what is the difference between them what is
 false positive what is false negative uh the 20:40 statistical test like the
 student T Test the F test Anova test uh the uh two 20:46 sample T Test the
 two sample normal test there are so many test that um it would 20:53 that um
 can be studied in this field of Statistics but there are a couple of of them
 that I uh selected and um I would 21:03 also provide you the links to that
 and you can also check them out and I would highly suggest you to study them
 also 21:10 this concept of P value is um very uh essential uh also this uh
 calculation of 21:17 the P value how you can use it how to interpret it its
 limitations and also this concept of 21:23 inferential statistics so blows
 like the central limit theorem the of large 21:29 numbers how it is used when
 it comes to this uh experiments and this is tied 21:34 back to the uh normal
 distribution function one of the most INF famous distribution function that
 you must know 21:40 as an AI engineer next up we have the dimension reduction
 techniques like the 21:46 principal component analysis or the factor analysis
 and you can also add here the panical correlation nysis so a 21:54 CCA so if
 you are looking for that one place that in organized way can help you 22:00
 to refresh your memory or to study all this in one place then you can also
 check out our fundamentals to statistics 22:07 course because we are covering
 there all these different topics which is a prerequisite and it's a must for
 you to 22:13 know before you get into the next level in your AI engineering
 Journey so once 22:19 you're comfortable with the mathematics and statistics
 you are ready to move on to the next step in your journey of 22:25 becoming
 an AI engineer the next skill set is the skills of data science so as 22:30
 an AI engineer you really need to have a good data science skills without
 good 22:36 data and without understanding whether you even have a good data
 or not and applying your data science skills um any 22:43 of other skills
 won't matter because um it's this phrase that is really uh easy 22:49 to
 remember you can have a great AI model but if you put a garbage in you 22:55
 will get a garbage out and that uh what you put into your AI model is your
 data 23:01 if your data is a trashy is a bad data and sometimes you don't
 even know that you are dealing with B data because you 23:07 don't have the
 data science skills then it doesn't matter how much effort or how much money
 you will put in your um AI 23:14 model how much gpus you will use or um how
 big your data will be if your data 23:20 quality is a bad one to understand
 these data skills you will need to have a data 23:26 science skills so what I
 mean by that so when it comes to um AI models they like 23:33 to work and
 they are performing good if they are dealing with the clean data your AI
 models also need to use a Data Science Skills 23:40 meaningful data a
 relevant one and also as an AI engineer you are responsible 23:45 for the um
 for the ethical side of your model and for that your data should be 23:52 uh
 unbiased as well so um as an AI engineer you will need to understand how
 23:58 to clean data how to Source data how to collect it if you don't have an
 AI engineer next to you and also how to 24:05 pre-process data and here I
 mean identifying the uh Missing data in your 24:12 database to understand
 what is the mechanism behind it is it missing a trandom is is it missing not
 a trandom 24:19 because this will then define whether you can impute the data
 so you can fill in this missing data what kind of 24:24 techniques you can
 use to fill in this missing data or maybe to drop it all together to
 understand whether you have 24:31 uh anomalies in your data outliers how you
 can use statistical and other techniques to find this outliers in your 24:38
 data and to remove it or maybe adjust it this concept of normalization you
 will 24:44 need to have a good understanding how you can filter your data how
 you can um 24:49 group your data um tell story about your data before you
 even get into the model 24:55 development section and how to uh split your
 data to have the skills of um 25:01 following the cycle of data preparation
 data evaluation and also using the data 25:07 as an input for your model
 whether it's a machine learning deep learning or an advanced generative AI
 model also 25:16 understanding how to uh visualize your data is really
 important as a data scientist you usually learn the um 25:23 exploratory data
 analysis and how you can use these different tools includ including Python
 and simple libraries 25:30 like Seaburn and metli to visualize your data and
 as a data science skill uh this 25:36 is a must to also identify outliers to
 identify certain Trends and also to tell 25:43 a story about your data so
 this is the basically the pre-work that you need 25:50 before you get into
 any moral development if you want to do everything properly and as a
 professional you also 25:56 need to understand uh Fe engineering skills which
 also is a data science skill so understanding how you can 26:03 create new
 variables so sometimes for example you have multiple variables but 26:09 it's
 not good enough because you just need one and it's usually a combination of
 this multiple variables and by 26:16 understanding how you can combine
 different variables in your database in one place and uh create one single
 26:22 variable is what we are referring as a feature engineering so you
 engineer the features that then you can use as an 26:30 input to your machine
 learning or your deep learning or your AI model in general so this is about
 the data 26:37 science knowing data signs uh will be um will set you for
 Success when it comes 26:44 to AI engineering career next up is the infamous
 traditional machine learning so 26:52 without understanding traditional
 machine learning there is no way to beable arounded AI engineer 26:58 um if
 you don't want to be in this position where for every single problem 27:04
 you use neural networks use you waste your company's money on the gpus or uh
 27:10 you spend a lot of time on using complex models that while you can use
 a simple 27:15 machine learning models if you don't understand this then you
 can never become this AI engineer that uh looks at 27:23 problems not just
 from a research perspective but also from business or Enterprise perspective
 27:28 so um that's why I always suggest to First Master the traditional
 machine 27:34 learning and then only get into the next point so here what I
 mean by traditional 27:40 machine learning I mean to um understand this
 concept of classification 27:46 regression supervised learning unsupervised
 learning these different algorithms that fall under these 27:53 categories
 like uh linear regression logistic regression decision trees uh 27:58 bagging
 boosting XG boost uh light GBM GBM and uh many other models including
 Traditional Machine Learning 28:05 unsupervised models like K means hierarchy
 Cloud string or DB scan in 28:11 which cases which of your models you can use
 the idea is that once a PM or a 28:16 business leader comes to you and tells
 you this vague business problem you as 28:22 an AI engineer you will need to
 quickly uh be able to figure out whether you are 28:28 dealing with a
 classification problem regression problem maybe an unsupervised learning
 program and you will also need 28:35 to have this uh quick understanding okay
 I'm going to use most likely this models 28:40 in order to solve that problem
 and being able to understand this will be really 28:47 important before you
 move on to any advanced moral uh studying so um Beyond understanding the
 28:55 algorithms and if I believe if I remember correctly those are about 23
 or 24 algorithms from traditional machine 29:02 learning understand their
 mathematics behind the statistics behind it what are their benefits what are
 their 29:07 disadvantages because in each of these categories you also need
 to understand how each of these models work and um 29:15 have this
 understanding that for this type of problems for example when you have a lot
 of missing data you can use 29:20 that model because it's more stable or if
 you are dealing with a data that follows normal distribution then you 29:26
 will then you can better use another type of model cuz for each of this
 classification regression or other type 29:32 of problems you will have many
 options and it's up to you as an AI engineer to 29:38 short list them and
 also from that to filter out which one you will use so beside this you also
 need to 29:45 understand how you can evaluate a tradition machine learning
 model what is this common cycle of the training 29:52 testing validation what
 are these different sampling techniques or resampling techniques uh what is
 29:58 bootstrapping what is cross viation what is kold cross viation or leave
 one out cross viation and also to understand 30:06 what are the different
 evaluation metrics depending on your problem you can use in order to evaluate
 your model 30:12 for example what is the difference between using the mean
 absolute um error 30:17 versus the mean squared error in which cases you can
 use which one or are or the root mean squared error or um how 30:25 you can
 evaluate a model that is in the field of classification it is the F1 30:30
 score um or it's the fbaa score which is more General version of the F1 score
 30:35 should you use recall should you pay more attention to the Precision Etc
 so uh understanding also when to use 30:43 machine learning when to use uh
 just rule based approach will be also important for you as an AI engineer so
 30:51 um that is about machine learning if you want to uh Master the field of
 machine 30:57 learning and everything that I just mentioned in one place you
 can also check out our fundamental to machine 31:02 learning course where we
 cover everything that you must know in order to become a well-rounded machine
 31:09 learning specialist you can also get a certification from lunatech Once
 you 31:15 complete your machine learning course so once you are comfortable
 with mathematics statistics and the 31:20 traditional machine learning next
 up is studying the Deep learning deep learning 31:25 is at the heart of the
 Modern Art artificial intelligence especially when it comes to generative AI
 so all these 31:32 different Cutting Edge tools like the chat gbt The Dol
 Sora or the um 31:39 different applications the um self driving cars the uh
 robots humanik 31:46 robots they are all based on narrow networks and narrow
 networks is this fundamental part when it comes to deep 31:53 learning think
 of the deep learning as more advanced machine learning where the 31:58 models
 are able to study better uh with a larger amount of data and this big 32:05
 data that uh the size of which increased more and more in the last decade
 made 32:12 the evolution of the deep learning more possible so when it comes
 to the Deep learning what I mean exactly is that you 32:19 need to understand
 how the Deep learning differs from the traditional machine learning you need
 to understand the 32:24 architecture of neural networks uh and how it works
 the concept of neurons the 32:30 perceptor this uh um in a simple way to be
 able to understand the structure of Deep Learning Foundations 32:36 neural
 networks the activation functions what it means this difference between
 different activation functions um and 32:43 also understand in which cases to
 use what this idea of hidden layers input 32:48 layer output layer um how
 they are related to the performance of neural 32:54 network um you also need
 to understand the concept of for forward PA backward pass the idea of B
 propagation what the 33:02 B propagation algorithm does the idea of loss function
 how you can calculate the loss function for a neural network also 33:09 how
 the training of neural network works so how it starts from the input then it
 goes to the forward path then does the 33:16 uh the loss calculation the back
 propagation Etc and also what is this 33:21 idea behind it and how using each
 of these different making each of these different decisions like the
 activation 33:28 function or the uh different optimization algorithms how it
 will be 33:34 impacting the performance of your deep learning model also
 understanding the different optimization algorithms like 33:40 the gradient
 descent stochastic gradient descent the RMS prop uh the momentum SGD 33:47
 Etc and of course the Adam or the adamw these different algorithms will be 33:53
 really important for you to understand how the Deep learning models are being
 trained and 33:58 optimize uh beside that you also need to understand the
 concept of Ving radiant problem the exploring radiant problem um 34:06 also
 understand um this different um computational graphs that are being used
 34:11 in order to represent uh NE networks um also um how you can evaluate
 the 34:17 performance of neural networks how you can use the cross entropy um
 and um 34:23 being able to understand these different um optimization
 technique makes the concept of mini badge gradient descent 34:30 is also
 important and the difference between bch gradient descent mini BGE gradient
 descent stochastic um gradient descent uh 34:38 understand the concept of
 Haitian uh why Haitian is is being used what it means 34:44 to have a faster
 versus better performing neural network um understand 34:50 also this batch
 normalization layer normalization what is the difference beside between them
 understand the 34:57 concept of residual connections and also what is uh
 gradient clipping cavier 35:02 initialization basically how you can
 initialize your neural network models of course um when I meant the
 fundamentals 35:09 of neural networks I definitely meant also understanding
 what is the bias what is the weights uh what it means to train 35:16 a neuron
 Network the role of improving these weights and also you need to 35:21
 understand the ways you can solve these different problems like how to solve
 a Venum gradient problem how to solve an 35:26 exploding gradient problem um
 and also um these different techniques to combat 35:33 the overfitting what
 it means to have an overfitting this comes from traditional machine learning
 but also in the Deep 35:38 learning it's still a problem and also understand
 how you

https://www.youtube.com/watch?v=nEwLBO8e0Dw

Intro to scrolling tabs in ChatGPT Atlas

OpenAI

https://www.youtube.com/@OpenAI

23-Oct-25

0:06 Hi there, my name
 is Darren. I'm an 0:07 engineer on the Atlas team. Uh today I 0:10 wanted to
 talk to you about a pretty 0:12 cool feature of the product. Um having 0:14
 to do with how tabs uh the tab system 0:17 works. So looking here, you can
 see my 0:19 browser after a pretty big busy day of 0:22 work. I've
 accumulated a bunch of tabs. 0:24 Um I have my calendar, my Gmail, Slack 0:27
 over here on the left. Um, when I'm 0:30 using these tabs that are pinned
 over 0:32 here, uh, when I use them, uh, and I 0:35 open links, links open
 near the pin 0:37 tabs. That's normal. If I want to maybe 0:40 do a search,
 I'm going to hit the plus 0:41 button here. Uh, tell me about the 0:44 latest
 features of Swift 6.2. 0:48 And um, my tabs opening on the right 0:50 hand
 side. I might do some other 0:52 searches. Tabs opening on the right hand
 0:54 side. I might go back to the tab I 0:56 opened over here. uh click
 around and 0:59 you know check out some different 1:00 things. Um maybe go
 back to the the 1:04 Slack, open some other tabs and you know 1:06 maybe this
 is a normal um thing that 1:09 you're used to in your browser. You have 1:10
 some tabs that are that you're working 1:12 with on the left, some tabs that
 you're 1:14 working with off to the right. Um you're 1:16 accumulating tabs
 over here, 1:17 accumulating tabs over there. Um 1:20 increasingly you're
 accumulating a 1:23 cluttering of tabs uh in the middle. and 1:25 many of
 these tabs in the middle, maybe 1:27 they're not as important to you anymore.
 1:29 So, you might take a moment and just 1:31 clear out a bunch of these
 tabs um so 1:33 that you can get back to a clean working 1:35 uh setup. Uh
 so, yeah, a lot of your new 1:38 tabs are on the right or on the left and
 1:40 it can feel pretty cluttered and 1:41 constraining. So, u we're aware of
 this 1:44 problem and this is a problem that's 1:45 been had been bugging me
 for a long 1:47 time. Started thinking about like how we 1:49 could
 potentially solve this. Um and so 1:51 we came up with a new system for uh
 the 1:54 tabs at the top that I want to tell you 1:55 about today. So in the
 settings uh 1:58 there's an option for for tab style 2:00 here. Uh classic
 tabs is the default. It 2:03 works the way probably used to your 2:05 browser
 working. We also have scrolling 2:07 tabs. When I enable scrolling tabs, the
 2:10 tabs suddenly change to be wider. Uh you 2:13 can see the title more
 easily on all the 2:15 tabs. Um, but importantly what you can 2:18 see is
 that uh you can see that the plus 2:21 button here is off to the left. I
 still 2:23 have my pin tabs, my calendar, my Gmail, 2:25 my Slack. And if I
 go into Slack and I 2:28 click a link, uh, it opens right next to 2:31 Slack
 like you'd expect. But if I open 2:33 if I want to do a search now, uh, it
 2:35 opens also on the left here. And as I do 2:38 a search or if I do other
 searches, 2:40 they're all opening on the left. as I 2:42 click other links
 in Gmail or Slack, 2:44 they're also opening on the left and all 2:46 the
 action, all the newer tabs are here 2:48 on the left instead of being both on
 the 2:50 left and off to the right. Um, and that 2:53 is really cool because
 it means the tabs 2:56 I'm working with stay together. Um, the 2:58 fact that
 they're not they're wider kind 3:00 of works because, you know, the set that
 3:02 I'm working with, I can see all of them. 3:04 I can still though get
 back to the older 3:06 tabs cuz they're off to the right. Here 3:08 I am
 scrolling with my touch a trackpad. 3:11 You can also scroll with the mouse
 wheel 3:12 and I can get back to some of these 3:14 older tabs. What ends up
 happening in 3:16 this system is that your older tabs get 3:18 sort of pushed
 off to the right and the 3:20 newer tabs are over here on the left. 3:22 And
 you know, I think this is kind of 3:23 cool. It means that um you can create
 3:25 keep creating tabs, keep working, 3:27 generating tabs um without really
 3:29 feeling like you have to stop and clean 3:30 things up. Um and that
 makes the whole 3:33 system just feel um a lot easier and 3:35 maybe a little
 bit less stressful when 3:37 you're working. So yeah, I really uh 3:40 really
 love this feature and I wanted to 3:41 share it with you. So I hope you get
 to 3:43 try it out and enjoy it too.

https://www.youtube.com/watch?v=8UWKxJbjriY

Introducing ChatGPT Atlas

OpenAI

https://www.youtube.com/@OpenAI

Streamed live on Oct 21, 2025

0:00 [Music] 0:05 Good
 morning. Today we're going to 0:06 launch ChatgPT Atlas, our new web 0:08
 browser. This is an AI powered web 0:10 browser built around chatbt. It's
 0:12 something we've been super excited about 0:13 and working hard on for a
 long time and 0:15 really excited to share with you today. 0:17 We think that
 AI represents like a rare 0:19 once a decade opportunity to rethink 0:21 what
 a browser can be about and how to 0:23 use one and how to sort of most 0:25
 productively and pleasantly use the web. 0:27 tabs were great, but we haven't
 seen a 0:28 lot of browser innovation since then. 0:30 So, we got very
 excited about the 0:32 opportunity to really rethink what this 0:34 what this
 could be. And in the same way 0:37 that for the previous way people used 0:38
 the internet, the uh URL bar of a 0:40 browser and the search box were a
 great 0:42 analog, the way that we hope people will 0:44 use the internet in
 the future and that 0:45 we're starting to see is that the chat 0:47
 experience and a web browser can be a 0:50 great analog. So, we got to work
 uh 0:53 designing a browser based around this 0:54 kind of experience. The
 browser is 0:56 already where a ton of work and sort of 0:58 life happens. And
 we think that by 1:01 having chatbt be sort of a core way to 1:04 help you
 use that that you can chat with 1:05 a page, you can use chatbt to find 1:07
 stuff. Um you can use an agent mode with 1:10 chut in a browser. Way more
 stuff that 1:12 we'll show you and you can try out 1:13 later. Um we can take
 this pretty far. 1:15 So we are excited to jump into a demo. 1:18 Um have
 some colleagues here. We'll 1:19 start with Ben for introductions and 1:20
 then we'll show you what we've got. 1:22 Great. Thanks Sam. Um I'm Ben. I
 lead 1:24 engineering for Atlas. So Atlas started 1:27 with a question. What
 if you could chat 1:29 with your browser? And from that idea, 1:32 we
 reimagined the entire experience, 1:34 replacing years of clutter and 1:36
 complexity with simple conversation. 1:39 We wanted to make sure that Atlas
 didn't 1:41 feel like your old browser uh just with 1:43 a chat button that
 was bolted on. Uh but 1:46 instead, we made chat GPT the beating 1:48 heart
 of Atlas. It's always by your side 1:50 and ready to help as you move across
 the 1:52 web. Uh, I find that when I use Atlas 1:54 myself, I'm more curious.
 I ask more 1:57 questions. I think it's made me just 1:58 like I said, a more
 curious, better 2:00 informed person. Um, we also made sure 2:04 that Atlas
 is fast and flexible enough 2:06 to support some amazing new experiences 2:08
 that we'll show you shortly. Uh, it's a 2:11 new kind of browser for the next
 era of 2:13 the web and we can't wait to show you 2:14 what it can do. So,
 Adam, do you want to 2:16 take us through some of the features? 2:17 Yes, my
 name is Adam, product lead for 2:19 Atlas. And as Sam and Ben mentioned a
 2:22 little bit about why we built Atlas, I'm 2:23 going to share a little
 bit about what 2:24 Atlas is. So first, Atlas should feel 2:27 very familiar.
 So it has all of your 2:29 tabs, bookmarks, autofill for password, 2:32 all
 the things you're used to. And then 2:34 there's three special core features
 of 2:36 Atlas that Ryan's going to walk you 2:37 through in a bit. The first
 is chat 2:39 comes with you anywhere as you go on the 2:41 web. So no longer
 do you have to copy 2:43 and paste between tabs when you're 2:45 working on
 writing an email or a 2:46 document. as you have that website up, 2:48 it'll
 just be right there for you if you 2:50 invoke it. And it'll have context of
 2:52 what you're working on, so it can be 2:53 more helpful. That's chat
 anywhere 2:55 across the web. The second big feature 2:57 is browser memory.
 And we talked a lot 2:59 about this when we were building it, but 3:01 memory
 is such a critical feature in 3:03 chatbt that people and users love today.
 3:06 And that's because as you use chatbt 3:07 more, it just gets more
 personalized and 3:09 helps you better and understands you 3:11 much better.
 Now, that's going to happen 3:13 as you go on your browser across the web
 3:15 in Atlas. and it just should be more 3:17 personalized and more helpful
 to you. 3:19 And then the third which we're really 3:20 excited about and uh
 Justin's going to 3:22 show this later is agent which is in 3:25 Atlas Chatbt
 now can take actions for 3:27 you. It can do things. So it'll actually 3:29
 bring up little cursor start clicking 3:31 around when you ask it to can help
 you 3:33 book reservations or flights or even 3:35 just edit a document that
 you're working 3:37 on. We're really excited to share this 3:39 with you. So
 Ryan, our lead designer in 3:41 the project is going to show you a tour 3:43
 of Atlas. Thanks, Adam. All right, so I 3:47 get to do the demo of the core
 flows in 3:49 Atlas. What you should see here is your 3:51 home screen. This
 is what you'll be 3:53 presented with when you first download 3:54 and open
 the app or anytime you create a 3:56 new tab. We tried to create an 3:58 experience
 here that will feel totally 4:00 familiar coming from a traditional 4:01
 browser, but with all the power of chat 4:03 GPT baked in. To that end,
 you'll see 4:05 there's a composer in the center of the 4:07 screen where you
 could ask chat a 4:08 question like normal. Can get to all of 4:10 your
 tools, 4:12 your models, 4:16 and your sidebar with all of your chat 4:18
 history. So, but because it's a browser, 4:21 you can do more. 4:25 Type
 hacker news. Chat's going to take 4:27 me to the URL. I could say I could
 4:30 reference a bookmark in human language. 4:34 and it's going to open my
 commits for 4:36 this galaxy diff. 4:38 You can use browser memory to search
 4:40 your web history for something that you 4:42 know you've seen before,
 but you don't 4:43 know exactly where it is. So, let me say 4:46 search web
 history for a doc about Atlas 4:52 core design. No, I made this somewhere.
 4:59 searching your browser memories. 5:05 There you go. Looks like it found
 the 5:06 doc I'm talking about. It's in my Google 5:08 Docs. If I tap it,
 you'll see it'll open 5:10 there. Let's jump back to the homepage 5:13 for
 one final feature. So, below the 5:16 composer on Atlas, you'll see 5:17
 suggestions. These suggestions are kind 5:19 of the first version of personalization
 5:21 in Atlas. Um, it will be generated for 5:24 you based on what Atlas
 understands 5:26 about what you've been up to or might be 5:27 trying to do
 next. They can be as simple 5:29 as a news story it thinks you might be 5:30
 interested in or as advanced as an agent 5:32 task that's going to delegate
 through 5:34 for you and uh and kind of click through 5:35 your tabs. Um the
 more you use Atlas, 5:39 the better these suggestions get. And 5:41 again,
 it's very much a vzero of 5:42 personalization, but we're really 5:43 excited
 to see where the homepage of the 5:45 browser goes as we um delve deeper onto
 5:48 this. Okay, so that's the home screen. 5:51 Now, I'm going to hop over
 to that 5:52 GitHub example and show you my personal 5:55 favorite feature.
 So, here I have some 5:57 code I was working on this morning. Um, 5:59 it's a
 shader for a little uh galaxy 6:02 generator. And in the top right, there's
 6:05 this ask chat GPT button. You'll see 6:07 this on any website you visit.
 And when 6:09 you click it, it creates a companion 6:10 sidebar. It's
 basically you inviting 6:13 chat GPT into your corner of the 6:15 internet.
 You can do all of the things 6:16 you'd expect to be able to do with 6:18
 ChatGpt, but now it can see whatever 6:20 that specific web page is. might
 sound 6:23 simple, but it's actually been a major 6:24 unlock for how I use
 the browser. It's 6:27 kind of gone from this tool that's very 6:28 much
 about displaying information for 6:31 you to edit into this tool that 6:33
 understands the information it's 6:35 displaying and in some cases can even
 6:36 edit it for you. So, it has a suggestion 6:39 here to just summarize the
 contents of 6:41 this diff. Let's ask for that and see 6:43 what it says.
 6:45 All right, it's a commit said even more 6:47 galaxy. It's updating a few
 of the 6:50 visuals and how this particle generator 6:52 works. This is cool.
 But what I really 6:54 want to know is, is this safe to 6:57 cherrypick into
 the R RC launching 7:02 today? 7:06 I thought we said no more changes today.
 7:08 There's always time for one more. Uh, 7:10 okay. Thinks this is pretty
 low risk. 7:12 I don't know about that. 7:13 Yeah, I'm not sure I totally
 agree with 7:14 that one, but it is just a visual 7:16 change. Um, and that's
 side chat. You 7:19 can use this in a wide variety of cases, 7:21 comparing
 products, bringing it into 7:22 your own corner of the internet. I use 7:24
 it a lot for pull requests or Slack when 7:26 I want to summarize a channel
 I've been 7:28 reading. Um, it's really useful and 7:30 we're excited for you
 all to try it. 7:31 I think also Ben mentioned how it makes 7:33 you more
 curious now that you have this 7:34 by your side. You just ask a lot more
 7:37 questions which I really love about it. 7:38 Totally. It's a little bit
 of a paradigm 7:39 shift where you go from just having this 7:41 sort of one
 call, one response to you 7:43 can kind of keep workshopping until you 7:45
 get what you're looking for, which is 7:46 very in keeping with chat. 7:48
 Yeah. I often find I'm browsing, I just 7:50 keep this thing open and I just
 like 7:51 flow questions into it as I go. 7:53 Totally. Speaking of keeping
 it open, 7:54 let's take a look at search, which has 7:56 uh some more of
 this side chat to show. 7:58 So, 8:03 I'm going to search for this movie 8:06
 I want to see. Um, and we've made some 8:08 major upgrades to search on chat
 gpt 8:11 when accessed via Atlas. So, we know 8:14 that um, search is kind of
 one of the 8:17 core flows in a browser for navigating 8:19 the internet. And
 a lot of these 8:20 searches can be very keyword- based or 8:22 short. Um,
 and LLM traditionally 8:24 struggle with that where they don't have 8:26
 enough context to provide a great 8:27 answer. So, one of the first things
 8:29 you'll notice is anytime you uh, search 8:31 within Atlas, you get these
 tabs across 8:33 the top. You can quickly pivot your 8:34 experience into
 something more like a 8:36 traditional search engine with images, 8:41
 videos, 8:43 or news stories, all without losing that 8:46 core chat
 experience on the home tab. 8:48 So, here, scroll down. Some nice images,
 8:51 a few uh updates on what this is. Let's 8:53 see if we can find a link.
 I'll take 8:55 this Roger Eert review. 8:58 It's given it four stars. One
 really 9:00 interesting thing here is that whenever 9:02 you click a link
 from a search result in 9:04 Atlas, by default, it's going to slide 9:07 chat
 over and open the web in a split 9:09 view. Now, if you don't want that, you
 9:10 can always commandclick the link or just 9:13 click the ask chat GPT
 button and close 9:15 it. But it has this kind of nice 9:16 property of you
 have a companion with 9:19 you as you search the internet. So, 9:21 maybe I
 want to go to a different review 9:23 here. I'll try this Yahoo one. 9:25
 Haven't you already seen this movie? 9:26 What's What's your review? 9:27
 I've seen it twice, actually. Uh, I 9:30 recommend it. Um, really really good
 9:32 actually. Um, let's just ask for a quick 9:34 summary of this review.
 Can you 9:37 summarize this review in five words or 9:41 less? 9:43 Maybe we
 can get to the meat of it. 9:45 This is where I think this this new 9:46
 model of search is actually really 9:48 powerful because it makes it it's
 like a 9:49 multi-turn experience. Like you can just 9:51 have this back and
 forth with your 9:53 search results rather than just being 9:54 sent off to a
 web page. You can use this 9:56 to really understand. 9:57 Totally. Yeah.
 9:58 Yeah, that's a great review, 9:59 huh? 9:59 PTA's best. 10:00 I have to
 check it out. 10:01 That's a high bar. Uh, definitely go it. 10:03 It
 honestly is great. Um, okay. Uh, for 10:07 the last demo I'm going to show
 you in 10:08 these core flows, I'm going to hop over 10:09 to my Gmail
 drafts. So, we know a really 10:12 popular flow in chat GPT is to draft 10:14
 some writing in a note or a doc or an 10:16 email. Copy that writing, bring
 it to 10:18 ChatGpt, workshop it a bit there, maybe 10:21 change the tone or
 tenor, um, language, 10:23 spell check, grammar, whatever it may 10:25 be.
 Get to something you're happy with. 10:27 copy the output of that, bring it
 back 10:29 to wherever you're working, paste it 10:30 there. With Atlas, we
 wanted to try to 10:32 flatten that flow into something that 10:34 feels uh
 like you can just do it in line 10:36 on any form field or text box on the
 10:38 internet. So, here I have an email was 10:41 writing to one of the
 other designers on 10:42 the team about this beautiful shader he 10:44 worked
 on for agent. I can just select 10:46 the text and hit the chat GPT nub.
 Maybe 10:48 I'll just say tidy my language. Doesn't 10:51 look like it was my
 best to begin with 10:54 back there. 10:56 Now I know why your emails are so
 10:59 polished of. 11:00 Yes. Well, uh um All right. There you 11:01 go. So,
 you get your update. I could ask 11:03 for another edit if I wanted. It lets
 11:04 you do all of this in line. Then when I 11:06 hit update, it's going to
 take whatever 11:07 your text selection was, replace it just 11:10 in that.
 It allows you perform really 11:11 scoped edits in a super useful way. We
 11:13 call it cursor chat. Really excited to 11:16 see what people do with
 it. Let's hit 11:17 send. Fire that off to Omar. 11:19 Awesome. 11:20 There
 we are. Those are the core flows 11:22 for ChatgPT Atlas. 11:24 That's
 awesome. Great work you guys. 11:25 Thanks very much. So that's a little bit
 11:27 about what makes uh chatbt in your 11:30 browser just an easier part of
 your 11:31 daily work. One thing that you can see a 11:34 little bit of there
 but really comes 11:35 through and use it is this is just a 11:36 great
 browser all around. It's smooth, 11:38 it's smooth, it's quick, it it's very
 11:40 nice to use. But now we want to show you 11:42 a more advanced feature
 um which is 11:44 agent mode in chatbt. Uh and so Pranov, 11:47 Justin, and
 Will are here to show you 11:49 that. Hey everybody, my name is Will 11:52
 Ellsworth and I'm the research lead for 11:54 the agent in Atlas. 11:55 My
 name is Justin. I'm an engineer on 11:56 the Atlas team. 11:57 And I'm
 Pranov, one of the product leads 11:59 on Atlas. 12:00 And we get to show you
 how Atlas is able 12:02 to browse the web and do things for you 12:05 in
 agent mode. 12:06 There's honestly so many different ways 12:08 you can use
 this, right? Uh maybe you 12:09 want to hand off a task that you're just
 12:11 not interested in doing or you want it 12:14 to teach you how to do
 something in 12:16 software you've never seen before. This 12:18 is a
 preview, but honestly, we've just 12:20 been blown away by how powerful this
 12:22 agent can be with full access to your 12:24 browser and your personal
 internet. 12:26 Uh, that makes safety really important, 12:28 right?
 Absolutely. And so, we've built 12:29 safety into every part of our stack
 from 12:31 the model all the way to the product 12:33 experience, which Panov
 will tell us a 12:34 bit more about. 12:36 But why don't we see it in action?
 12:37 Let's rock and roll. 12:38 All right. So, we have been planning a 12:41
 haunted house. 12:42 Really excited for this. 12:43 Yeah, I'm I'm pumped. And
 uh for 12:45 whatever reason, I got roped into being 12:47 the project
 manager for this. And uh we 12:50 have a Google doc that we've been using
 12:52 to kind of informally plan out our 12:54 tasks. And so you can see um
 you know 12:56 some people have filled in their current 12:58 week's tasks.
 And uh unfortunately there 13:00 are a couple of issues here. So the 13:02
 first problem is as you can see by the 13:05 to-dos uh some people have not
 13:10 uh filled in their current week's task. 13:12 uh and so I would love to
 leave a 13:14 comment politely reminding them to do 13:15 so. And then second
 is while Google Docs 13:18 is this amazing tool uh we also have 13:20 some
 more formal task management 13:22 software called linear and I would love
 13:25 to take all the current week tasks that 13:27 have been filled out and
 convert them 13:29 into linear tasks or uh in the linear 13:31 verbiage
 issues. So the tough part here 13:34 is I have very little project management
 13:36 experience. Don't really know how to use 13:38 linear. 13:38 I don't
 know why we put you in charge of 13:40 this. 13:40 Yeah. uh beats me. But um
 I therefore 13:44 would love to just delegate this uh to 13:46 agent mode in
 Atlas and have it take 13:48 care of this for me. And so what I can 13:50 do
 is I can click uh this agent mode 13:52 here. And you can just find this with
 13:54 the the plus button selecting agent 13:55 mode. And I'm going to kick
 this off. 13:57 And this agent mode tells Chat GBT that 14:00 I want it to
 actually take actions on my 14:02 behalf inside of Atlas. And so you see
 14:04 it has its own cursor. It's going to be 14:06 clicking around as if it
 were me. has 14:08 has access to all of my local 14:10 authentication, all of
 my history. Um, 14:12 it should really feel like a natural 14:13 extension of
 myself. And I'm going to 14:15 hand off over to Justin. 14:16 Yeah. Yeah. The
 team paid a lot of 14:18 attention to the product experience 14:19 here,
 right? We really wanted to make it 14:21 feel like it was coming alive. You
 could 14:22 see exactly what the agent was doing. 14:24 So, you could start
 to build trust that 14:25 it was, you know, doing what you wanted 14:26 it
 to. 14:27 But yeah, just just to emphasize this 14:28 point, this is Chpt in
 agent mode using 14:31 your web browser for you locally. It's 14:33 got all
 your stuff. It's clicking around 14:34 for you. You can watch it or you
 can't. 14:36 You don't have to, but this is like 14:37 really it's using the
 internet for you. 14:39 Exactly. Exactly. 14:40 Yeah. It's like right in your
 tab. And 14:43 that's one of the cool things about the 14:44 experience of
 using agent 14:46 in Atlas. 14:48 So, it looks like it is kicking off. So,
 14:51 one thing that's really nice is that I 14:52 don't need to sit and
 watch it, right? I 14:53 can let it just do its thing in the 14:55 background
 14:56 um and use my browser for other things. 14:59 So, here we have a
 recipe. We're uh 15:01 we're planning a potluck, right? 15:02 Yeah. Really
 excited about this recipe. 15:04 Yeah. Yeah. So, I'd like to show you how
 15:06 we can use agent for things in in in 15:08 your personal life. So, one
 thing that I 15:11 always struggle with with recipes is 15:12 figuring out
 what ingredients I need to 15:14 buy, right? Uh it's somewhere in the 15:16
 recipe page. It's some serving size. I 15:18 need to figure it all out. So, I
 I like 15:20 to use Atlas to ask Chat GPT, uh what 15:24 ingredients 15:26 do
 I need to buy to cook do I need to 15:30 cook this for eight people? 15:34
 and Chacht is going to go ahead and read 15:37 the web page, figure out the
 15:38 ingredients, kind of do some math for 15:41 me, and tell me exactly
 what I need. 15:43 So useful. 15:44 Yeah, in the past, I've told it that I
 15:45 like my uh I like my shopping list 15:47 organized by grocery aisle to
 make it a 15:49 little easier to shop for. 15:52 And looking at this, you
 know, I have 15:54 most of this, honestly. I just need the 15:55 meat and the
 produce. So, I'm going to 15:56 say, uh, can you order the meat and 16:02
 produce for me? and we'll shut off how 16:05 you can start agent mode by
 clicking a 16:07 button, right? Which is really useful if 16:08 you know to
 reach for it. But in those 16:10 moments that you don't, chatbt can 16:12
 figure out that the way to accomplish 16:14 this is to take over your
 browser, 16:16 right? Uh you're always in control. You 16:18 always have the
 option to approve or 16:20 reject it. So I'm just going to click 16:22
 continue uh to hand hand the task off to 16:24 agent. 16:25 Yeah. And I I
 love how collaborative 16:27 agent is in Atlas. So you can just hand 16:30
 off your tabs, you can go back and 16:32 forth. And we've really improved
 agent a 16:34 bunch to make sure that it's a lot 16:36 better and faster at
 these collaborative 16:39 tasks. And as you can notice, like at 16:41 any
 moment in time, you could take 16:43 control. And so one thing that's really
 16:46 great about this is like agent already 16:48 knows that Justin likes to
 shop at 16:50 Safeway on Instacart. And so it knows 16:53 exactly where to go
 when all he said 16:54 was, "Can you order this for me?" And so
 16:56 it's found its way over to Instacart. 16:58 and it's starting to
 search. You can see 17:01 how it like types way faster than I do. 17:04 Um,
 and 17:05 I pride myself in my typing speed and 17:07 this has just blown me
 out of the water. 17:10 Exactly. And it started adding items to 17:11 the
 cart already. And so, uh, I want to 17:14 take this moment actually to to
 talk 17:16 about, um, you know, despite all of the 17:18 power and awesome
 capabilities that you 17:21 get with sharing your browser with 17:23 ChatGpt,
 that also poses an entirely new 17:26 set of risks. And so it's really 17:28
 important to us in addition to a bunch 17:30 of built-in safeguards like chat
 GBT 17:32 agent is only ever operating on your 17:34 tabs. It can't execute
 code on your 17:36 computer or access other files. It's 17:39 just in your
 tabs that you're also in 17:41 control of exactly what you're handing 17:42
 over access to. And so if I open a new 17:44 tab just to show this off, you
 always 17:46 get to decide whether chat GBT agent is 17:49 logged in or
 logged out. And so we 17:50 really recommend thinking carefully 17:52 about
 for any given task, does chat GPT 17:55 agent need access to your logged in
 17:57 sites and data or can it actually work 18:00 just fine while being
 logged out with 18:02 minimal access? And that same principle 18:04 of
 control carries through to our entire 18:06 browser experience. Ryan showed
 off 18:08 these awesome uh browser memories that 18:10 power these
 suggestions earlier. It's 18:13 it's also worth noting that those are 18:14
 completely optional. You can decide 18:16 whether you turn them on in
 onboarding 18:18 or not. you can always see the memories 18:19 themselves and
 manage them in settings. 18:22 And for anytime you don't want um uh you 18:26
 don't want this to be remembered by 18:30 chatgpt 18:31 uh you always can
 make a new incognito 18:33 window. And so you'll be able to do this 18:36 to
 ask questions like what to do when 18:40 your palms are sweaty on a live
 stream. 18:45 Asking for a friend, right? Yeah, of 18:47 course. And I'm
 realizing I don't think 18:50 I want everyone to see the answer to 18:51
 that. So, why don't we go back and 18:53 check? 18:53 I don't know if I need
 you using my 18:54 computer either. Okay, great. 18:56 Should we go back and
 check how the how 18:58 the task went? 18:58 Let's do it. 18:59 So, here's
 our Instacart order. Awesome. 19:01 You can see that in just about two 19:02
 minutes, the agent was able to go 19:04 through, fill out the cart, and it's
 19:06 just so useful having um the cart filled 19:09 out and delivered to you
 like this, 19:10 right? It doesn't need to go all the way 19:11 to making the
 purchase order. In fact, 19:13 it's better for me if I can review what 19:15
 it did and decide to buy or add more 19:18 things to my cart or whatever else
 I 19:19 need to do. 19:20 Yeah, 100%. 19:21 Cool. And then let's take a quick
 look 19:23 at the linear task. 19:26 Um, and so yeah, looks like it 19:28
 successfully added these tasks to 19:30 linear. And it's a little hard to see
 on 19:32 the screen, but it's also tagged the 19:34 right people for each
 task. 19:36 One cool feature is it shows you 19:38 relevant tabs at the
 bottom, so you can 19:39 see what tabs it's worked on. So, I can 19:41 go
 back and check the Google doc and 19:43 see. Great. It looks like it's tagged
 19:45 all the people uh who had the to-dos and 19:47 given them a plight
 reminder to fill 19:49 this out. 19:49 It's going to save me so much time.
 19:51 Yeah. Um and and save my job because I 19:54 was not uh familiar with
 this whole 19:56 project management thing. 19:57 So, uh we've seen a couple
 of awesome 20:00 examples of how chat GBT can actually 20:02 control the
 Atlas browser and perform 20:04 useful actions on your behalf. And so in
 20:06 the same way that GBT5 and Codeex are 20:09 these great tools for vibe
 coding, we 20:11 believe that we can start in the long 20:13 run to have an
 amazing tool for vibe 20:15 lifing. So delegating all kinds of tasks 20:18 uh
 both in your personal and 20:20 professional life to the agent in Atlas.
 20:23 You know, one of the great joys of 20:24 working at OpenAI is when we
 release 20:26 technology, people outside the company 20:28 always come up
 with way more creative 20:30 ideas for how to use it than we can. uh 20:32
 maybe we're just not uh super creative 20:34 folks, but I'm really excited to
 see all 20:36 the unexpected and cool ways that you 20:38 can use the agent
 in Atlas and we're 20:40 really excited to ship this. So, with 20:42 that,
 back to Sam. 20:44 We are indeed really excited to ship 20:45 this. We we
 hope you'll love it. So, 20:47 this is going live today for Mac OS 20:49
 worldwide uh for all of our users, 20:51 although agent mode is only
 available to 20:53 plus and pro users for now. We want to 20:55 bring this to
 Windows and to mobile 20:56 devices as quickly as we can. We think 20:58
 people will uh hopefully will you'll 21:00 love this as m much much as we do.
 21:03 There's a lot more to add. This is still 21:04 early days for for this
 project. We we 21:07 think we the kind of idea that we're 21:09 excited about
 is what it means to have 21:11 custom instructions follow you 21:12
 everywhere on the web. And as you have 21:14 this agent that you're having do
 things 21:15 for you, getting to know you more and 21:17 more, pulling stuff
 together for you 21:18 proactively, finding things that you 21:20 might want
 on the internet and bringing 21:21 them together, which we we showed a 21:22
 little bit of. We think we can push that 21:23 quite far. So, we hope you'll
 check this 21:25 out. We hope you will uh enjoy it. and 21:27 we please send
 us feedback. Thank you 21:29 very much.

https://www.youtube.com/watch?v=tl4ke1EeFVE

The Story of Apify | 10th Anniversary

Apify

https://www.youtube.com/@Apify

20-Oct-25

0:03 Apify started with
 an accident ten years ago 0:06 The founder clicked submit instead of save
 draft on his 0:09 YC Fellowship application. Full three days before the
 deadline 0:12 Reading the submission 10 years later, the plan is still valid
 0:16 Apifier will be a cloud service for developers 0:18 to turn any website
 into an API, which will enable them to rapidly 0:22 build new apps on top of
 existing third-party web apps and data sources 0:26 Back then, he realized
 0:28 companies need ever more data. And the web is the largest source of it
 0:31 In 2015, it was just him and co-founder Jakub Balada, two young computer
 0:36 science graduates 0:37 So they built a new kind of web crawler that made
 it easy to get that data 0:41 With this project, they applied to the YC
 Fellowship 0:43 a 10,000 kilometer flight, a ten minute interview, 0:46 and
 somehow they got in. 0:47 What followed were two months of the most intensive
 work of their lives 0:51 The goal was to turn this idea into a real product
 0:56 Then on October 20, Apifier launched on Hacker News. 1:00 Thousands
 tried the free demo, but what mattered most were the first 1:03 120 users
 brave enough to leave their email 1:06 They weren't just early adopters, they
 were the seed of a community 1:09 Soon after, the company found investors who
 believed in them, started growing 1:13 the team, 1:14 the revenue, and the
 office space 1:15 The next turning point came in 2017, when the young startup
 introduced Actors 1:20 A new way to package, run, and sell software services
 in the cloud. 1:24 A tool became a platform 1:26 and Apifier 1:28 became
 Apify 1:30 Customers noticed, 1:32 and from there, the real growth started
 1:34 In 2020, Apify Store opened to the public, who could start selling their
 Actors 1:39 Today, it hosts over 7,000 Actors for all use cases 1:42
 imaginable, and pays out the Apify creator community 1:45 half a million
 dollars per month 1:47 We started 10 years ago as two developers solving our
 own problem, 1:51 and somehow we?€?ve built the world?€?s 1:52 most vibrant
 community and marketplace for web automation tools 1:56 And a company 1:57
 that I still very much enjoy working for, full of great, smart, and fun
 people 2:00 Building and selling software changed our lives, 2:03 and now
 we?€?re helping people around the world 2:05 let it change their lives too
 2:06 Ten years ago, we set out to make the web more programmable 2:10 Ten
 years from now, 2:11 Apify will be the world?€?s largest marketplace of 2:13
 AI tools, enabling anyone or anything to get more value from the web 2:17 The
 future is clear 2:18 The question is 2:20 Will you join us in building
 it?

ビデオのメタデータとコメントの抽出

チャンネルからのビデオリストのスクレイピング

YouTubeビデオURL

はい

スクレイピングするYouTubeビデオのURLを1つまたは複数入力します。改行区切りの入力に対応しています。

https://www.youtube.com/watch?v=Ri-HcFlNcJk