Every New ChatGPT Feature Explained & More AI News You Can Use

0:00 Welcome to another week of AI innovation.

0:02 And this week is one of those that I love

0:03 to do because we have a bunch of ChatGPT updates.

0:06 And I happen to know that most of you

0:07 are still using ChatGPT as their main used app.

0:10 And let me just say a sentence about the fact that that is the case for me, too.

0:14 70 to 80% of my personal AI usage is within ChatGPT or Claude.

0:19 I play in all of these other apps, and there certainly uses,

0:22 but as somebody who doesn't rely upon coding on a day-to-day,

0:25 for me roughly 2/3 of my work still happen in ChatGPT or Claude.

0:29 Now, mind you, I say work.

0:30 I experiment and play a lot in the others.

0:32 And today, we have a bunch of practical ChatGPT updates for you,

0:36 along with updates from Google and Claude.

0:37 The big boys all released a bunch of interesting things this week.

0:40 We're going to talk about all that and more

0:41 in this week's episode of AI News You Can Use,

0:43 the show that pulls together all the AI innovations from across the internet.

0:46 We filter for the stuff that matters and is usable,

0:49 and then I have the pleasure of presenting it back to you.

0:51 Let's begin.

0:52 The big story this week is the story of OpenAI.

0:54 They're raising a brand new round,

0:57 and I guess to argue that and get some public attention,

1:00 they also released a bunch of features

1:02 inside of their core product along with that.

1:04 They raised $122 billion at a $852 billion valuation.

1:09 That's almost a trillion dollar valuation for a private company.

1:13 I believe that's the single biggest fundraise in venture history.

1:17 And all of that happens with this mission to build a unified AI super app.

1:21 Now, before we get into the features,

1:22 I just want to give you a little bit of context because,

1:24 as it has been recently revealed, most of ChatGPT's revenue,

1:27 around 3/3 of it, come from ChatGPT subscriptions.

1:31 Whereas for company like Claude, it's actually pretty much the opposite,

1:34 where most of their revenue comes from API usage,

1:37 developers and other applications using it.

1:40 That also includes Claude code and open Claude, by the way.

1:43 And now that OpenAI clearly stated

1:44 that their goal is building this AI super app,

1:47 what they're referring to is expanding their consumer-facing

1:49 application and really making this beautiful chat interface

1:53 that we're all so familiar with into something

1:56 that is AGI or the one unified application including everything.

2:00 And I guess to prove to the public and to the investors,

2:03 now we can get into the features, they released a bunch of stuff this week.

2:06 I'll start with my personal favorite,

2:08 which is the ability to share your location with ChatGPT,

2:10 and ah yes, yes, you know, privacy concerns, I know.

2:14 It is very convenient, and I'll show you that in a second,

2:17 but first let's look at how to access it.

2:20 Go to your settings in ChatGPT, data controls,

2:23 and now you have this location feature.

2:25 This is brand new as of this week,

2:26 and they didn't even have a proper announcement around this.

2:28 I find this so convenient though,

2:29 because when I have this on, I can ask questions like,

2:32 what is a good co-working cafe near me?

2:35 This will also work on the phone, of course,

2:38 and now that it knows I'm in Miami right now,

2:40 it can make all of those recommendations.

2:42 And as you can see, this is pretty much my location right here,

2:45 so I can use the interactive app here.

2:47 There's beautiful pictures in here.

2:48 Notice how in previous episodes I was saying they're really

2:50 moving away from this wall of text at all means.

2:53 And even a simple search like this like has maps, images, links.

2:57 This is that direction of moving towards

2:59 this AI super app away from just a chatbot.

3:01 So yeah, this one looks really nice, and I might check that out,

3:04 and can use it for anything you do.

3:05 It just makes it convenient,

3:06 and maybe even superior to using like a Google Maps to finding these things.

3:10 I mean, the mostly fake reviews on there aren't worth much.

3:13 To me personally, this way of presenting it with the map and then

3:16 the results and the images is just preferable to only having the map and then,

3:20 you know, having to open each entry.

3:22 And then finally, I would add, on the phone you can set so that it

3:24 can only use a location while you're in the app.

3:26 That might be the smart thing to do.

3:27 It's a very convenient thing that Gemini already had, and now Chad has it, too.

3:31 Okay, so let's look at some other changes.

3:32 A big one is that if you have a lot of text and you add it to ChatGPT,

3:35 so let's say I just follow up my first

3:37 location chat here with expand on each one of these.

3:40 That's the thing I like to use.

3:41 I also like to use the opposite of give you

3:43 a more concise version when I need to get the high level.

3:46 And the novel feature here is that if

3:47 you give it text with over 5,000 characters,

3:51 it won't just add it as text to a new chat or an existing chat anymore.

3:54 It will add it as an attachment.

3:56 There's a difference there.

3:57 And I need you to be aware of this difference to get the most out of this tool.

4:00 So, what I'll do is I'll just select all and copy that.

4:02 And if I go into a new chat,

4:04 you will see that if I add it, it's not added as text.

4:07 It adds it in a .txt file, as you can see here.

4:10 Now, here is the deal.

4:11 Attaching something isn't equal to adding it as text.

4:15 That's why they give you this button here, show in text field.

4:18 So, if I press show in text field,

4:19 it will actually appear in my chat box right here.

4:22 And all of a sudden, it's here.

4:23 Now, you might be asking, "Igor,

4:24 what's the difference between attaching it and having it in here?" Well,

4:27 they were never super clear on this.

4:29 But if it's as pure text in the box, as you can see it right here,

4:33 every single detail is added to the context

4:36 window that you're limited by within ChatGPT.

4:38 Just as a quick refresher, every AI model has a so-called context window,

4:42 which you can think of as a certain limitation on how much it remembers.

4:45 It's basically a rolling window.

4:46 So, if you give it more input than it has context,

4:48 then the context window will roll down and forget what's in the beginning,

4:51 and AI basically gets amnesia.

4:53 Now, what happens if you add something as an attachment

4:56 is that it won't be directly added to the context window.

4:58 There's some sort of summarization step

5:01 in between this document and it adding it

5:05 to your prompt as you would if you're just writing in here like so.

5:09 In other words, in attachments, you sometimes tend to lose certain details.

5:13 Now, not always, but if you really wanted to retain every detail,

5:16 adding it as text versus an attachment is superior,

5:20 but it will limit the amount of other context you can add.

5:23 Now, if you don't believe me, they even highlighted this in the documentation

5:26 that they never promoted publicly, though.

5:28 Yeah, if the text is over 5K characters,

5:30 ChatGPT will convert it into an attachment.

5:32 This keeps the composer cleaner.

5:34 This is the composer right here.

5:35 And, this is the critical part,

5:37 helps prevent large pastes from consuming your entire context window.

5:42 As a user, I want it to consume my context window.

5:44 They as a company don't want us to consume the context

5:47 window cuz they have to pay for all of those tokens,

5:49 for all that context running through their system,

5:51 especially if you continue to chat.

5:52 They'll have to run all of those tokens

5:54 through the system again and again and again,

5:56 and you're just paying for $20 subscription.

5:58 So, they don't want that.

5:59 We as consumers usually do, unless you really have a lot and you don't

6:02 want it to be consumed by the context window.

6:04 So, my recommendation would be if you're doing

6:06 quick work and you add text in there,

6:08 then press this button and you'll get more

6:10 precise recall on what you actually added in there.

6:13 Good to know.

6:13 Few more ChatGPT updates here.

6:14 And these are quick ones cuz they're

6:16 just single features that most people aren't using.

6:18 For example, Google Drive had multiple apps.

6:20 So, you had Google Docs, you had Google Sheets,

6:22 you had Google Slides, and then you also had Google Drive.

6:24 Now, they unified it into just Google Drive.

6:26 When you connect Google Drive, gosh,

6:28 if I say that one more time, I'll have a seizure.

6:29 If you connect the app that I just mentioned so many times,

6:32 it will also connect Google Docs and all the others.

6:34 If you go to my account settings and under apps,

6:36 you will see that it's all unified here,

6:39 and you might have to click enable sync once to reactivate everything.

6:42 If you connected the others before,

6:43 you don't have to worry about this, it's all connected.

6:45 But, if you're using this regularly,

6:46 I would actually go in here and make sure it's reconnected,

6:50 so it's all clean under one connector.

6:52 If you're watching the show,

6:53 you'll know I'm not a big fan of these connectors cuz they're very leaky.

6:56 They do some of this like summarization in between

6:58 bringing in stuff and putting it in here.

7:00 So, you can't really fully rely upon it.

7:01 It can be useful though,

7:02 especially if you have shared context with other teammates.

7:05 Anyway, it's simpler now, all on the one connector.

7:07 It's a step in the right direction towards having a AI super app.

7:10 Hmm.

7:11 And as we're talking about apps, they actually updated some of them.

7:14 So, they have updates to Box, Notion, Linear, and Dropbox.

7:17 And basically, you can just do more stuff

7:18 with the apps within ChatGPT, which can be nice.

7:20 I don't know about you.

7:21 Please leave a comment if you are,

7:22 but me personally, I don't really use these much.

7:25 I just find them to always do the first step in the workflow inside of the app,

7:28 and then I have to go out and go into that app anyway.

7:30 Might as well do that immediately rather than doing one

7:33 step in chat and the next three steps in Dropbox.

7:36 Anyway, if you use them,

7:37 you can now upgrade them now to the new versions like so.

7:40 They also made some updates to the mobile application,

7:42 so if you check that out here, basically created a full screen view of the menu

7:46 and you can kind of swipe to it, which is nice.

7:48 Everything is larger here.

7:49 You have your chat history, GPTs, image generation, or the apps on top.

7:53 I like it.

7:54 Basically, there's two screens now for menu and chat.

7:56 Another quick ChatGPT change, and I'm excited for this in my car,

8:00 is ChatGPT voice mode in CarPlay.

8:03 So, you can, you know, talk to the assistant via CarPlay.

8:06 I wonder how deep the integration runs with other apps.

8:09 We have a really nice if it would actually act as a Siri that works for my car.

8:14 I like that idea.

8:15 You need to have iOS 26.4 or later and you

8:19 need to have a iPhone and iOS running that supports CarPlay.

8:22 Another thing that they improved is the shopping experience within chat.

8:25 Remember they introduced partnerships with several brands and now

8:27 you could see the product directly inside of chat.

8:30 Interface wasn't the best.

8:31 They overhauled it now and actually now it looks fantastic,

8:34 as you can see right here.

8:35 It also looks at the shops more frequently,

8:37 so it has the freshest products from the various providers.

8:39 And then finally, they added some plugins and codex,

8:42 if you use that instead of cloud code,

8:44 which I don't know many people if anybody who do that.

8:46 Everybody that I know is in cloud code or open

8:49 cloud and they're messing around with that type of stuff.

8:50 If you are using codex, there is new plugins and that is pretty much it.

8:54 Updates on the road to the super app.

8:57 Interesting.

8:57 We'll keep you posted every week here.

8:59 Let's see what else we have this week.

9:00 There's a few more interesting ones, starting with the Google releases.

9:03 So, they added a bunch of new models and usually new model releases,

9:06 hmm, not that interesting,

9:08 but Gemini 3.1 flash live brings kind of a new experience to using

9:12 Gemini because it's so darn fast and it works with voice mode.

9:16 So, basically, this is an upgrade to ChatGPT's advanced voice mode, you know,

9:20 the mode where it talks to you and you talk back and it talks back again,

9:23 but inside of Gemini and now it's faster than ever with a larger context window,

9:27 so it remembers more.

9:28 But here on my phone, there's a button and now I can talk to Gemini.

9:32 Hey, what up?

9:32 Give me a comprehensive intro of yourself and what you can do for me.

9:35 Sure, I'm a large language model trained by Google.

9:39 I can understand and generate text.

9:42 can I interrupt you easily?

9:44 Okay, that was pretty good, not bad.

9:45 Now, respond instantly.

9:52 You bet.

9:52 I'm ready when you are and I'll do my best to keep the pace going.

9:56 Okay, great.

9:57 So, basically, I can share video and stuff like that, too, right?

10:00 So, if I give video access Hey, what's up?

10:02 I'm just filming right here.

10:04 What kind of setup is this?

10:05 Uh what platform do you think I'm recording for?

10:08 That's a pretty professional setup.

10:10 I see a microphone, camera, and lighting, plus a script or notes on your laptop.

10:16 The layout suggests you might be recording

10:18 for a YouTube channel, online course, or

10:21 Nice.

10:22 It got that well.

10:22 So, it is snappier, definitely faster than the previous one,

10:25 very comparable to ChatGPT,

10:27 but with the various features like sharing your screen,

10:29 sharing a camera, it being fast, larger context window.

10:32 But, they've been kind of quiet about this category, right?

10:34 We had all of these fancy demos out of ChatGPT with the voice mode.

10:37 They did all the fancy demos out of Google about the voice mode,

10:39 but the stuff didn't really ship.

10:41 But, now it's coming together with the new model,

10:43 ability to share video, and all that stuff, larger context windows.

10:46 Gosh, it almost seems like Google is also trying to build AI super app.

10:51 Who would have thought?

10:51 Who would have thought?

10:52 Not me.

10:54 So, that's Gemini 1.5 Flash live.

10:56 Now, also, they released a family of open-source models,

10:59 which I'm actually becoming increasingly interested in because,

11:01 by the way, I have this new MacBook here.

11:03 And this is me laughing because, you know,

11:05 the broke 10-year-old kid living with a single mom, moving around all the time,

11:09 kind of looks at this and is like,

11:10 "No way." But, it's the maxed-out one and I don't say that to brag.

11:14 I say that to show you what you can get for $8,000 today.

11:18 Apple M5 Max, 128 GB, and it can basically run 200 billion parameter models.

11:24 It's super slow with models like Kimmy, but if you run something like 1.5,

11:29 gosh, it's faster than ChatGPT and the quality of it is actually close to chat.

11:34 Anyway, I just briefly wanted to talk about that.

11:35 Maybe I'll make a dedicated video about local models and this machine soon.

11:39 If you guys are interested in that, leave a comment if you are,

11:41 by the way, or leave a thumbs up on a comment that says it is.

11:44 It's kind of hard to gauge if people actually want to hear about that stuff.

11:46 Anyway, Google released new open-source models, which could run on this machine.

11:50 That's why I pointed this out.

11:52 The models are the Gemma family of models.

11:55 And I don't know much about these yet.

11:56 Usually takes a few days for people to run their benchmarks

11:58 and to get experience with these to make a judgement.

12:01 But from the benchmarks, gosh,

12:03 even the 31B models seem as capable as the DeepSeek

12:07 3.2 models and almost as capable as the Kimiko models,

12:11 which for many people is the golden standard

12:12 for running them with their open claw Mac minis.

12:16 So yeah, Google threw their hat in the ring.

12:17 They have new models.

12:18 Really impressive scores.

12:19 If you use offline models, look into them.

12:21 Okay, and here's a bunch of quick stories that I wanted to mention,

12:23 but maybe we don't need a full segment for them.

12:25 So there's two music generators that came out this week that were updated.

12:28 One of them is Suno 5.5.

12:30 And honestly, I'm you know, I think Suno is fantastic.

12:32 It creates probably the best quality songs in the market,

12:35 but the latest release is all sort of sound the same to me.

12:39 And other people seem to be sharing this opinion.

12:41 They got so good previously that it's kind of hard to top it.

12:44 It's similar to the Midjourney situation.

12:45 Cool new release, but the bar was so high

12:47 that the new bar is kind of at the same level.

12:49 Their main focus was more personalization and user control.

12:52 So if you use that tool, check out the new version of it.

12:54 Google has their own horse in this race.

12:55 It's called Lyria, and they updated the newest Lyria,

12:57 Lyria free pro, that can create three-minute tracks.

13:01 In terms of quality, it's definitely the main competitor to Suno.

13:03 The fun thing is they're starting to integrate this into other tools.

13:07 For example, Google's Gemini.

13:08 You can use it right in Gemini.

13:09 You can ask for a track and it just creates it.

13:11 Which again, AI super app, to bring everything together.

13:14 That's what's happening here.

13:15 Here's a quick snippet of a track created in Lyria.

13:18 Trading the ocean for frozen highs.

13:21 Slide the rail, feel the wind under the flipper.

13:24 Slide the rail, feel the wind under the flipper.

13:27 Slickest soul, baby, cold as zipper.

13:30 Compared to the same track created in Suno.

13:42 Okay, next there's a Claude code update where

13:44 they give it the ability to use your computer.

13:46 Yep, basically it can click buttons on your computer now

13:49 which goes beyond what it was able to do before.

13:51 Huh, almost seems like they're also gunning for this idea

13:53 of an AI super app but at a deeper,

13:56 way less user-friendly level than ChatGPT is.

13:59 So that's that.

13:59 Usually computer use is still so flimsy that you can't fully rely upon it.

14:03 And also usually it works well with things like sheets.

14:05 But then you do have to add a loop for it

14:07 to double-check what it did which also doesn't always work.

14:10 Worth the experimentation and the aches in some cases.

14:12 Then there was a leak around this Claude model called Mythos, Myth, Mythos.

14:17 It's one of those two pronunciations.

14:18 And I don't even know.

14:19 Today some people were saying it was just an early April Fools joke

14:23 cuz there was also a Claude code leak that was April Fools joke.

14:26 I'm not sure.

14:27 I think they are definitely working on another model.

14:29 This is supposed to be the successor to Opus 4.6, their best model right now.

14:33 It's basically just a rumor but I wouldn't be surprised if we see a new

14:35 flagship Claude model within the next month which this was supposed to be one.

14:39 There's also Claude tools on mobile.

14:41 We test these.

14:42 Wanted to create a dedicated video around these.

14:44 But from testing them,

14:45 we arrived at the same conclusion that we get every time we try

14:48 these connectors and tools and apps and it's

14:50 just the fact that they're not reliable.

14:52 They get the job done once, second time something happens,

14:55 then it works again, then it doesn't, then it forgets something.

14:57 It's just like working with a person that every time you tell them

15:00 something there's like a 40% chance

15:02 of them forgetting it or doing something different.

15:05 That's no fun and neither are these Claude tools on mobile.

15:08 Can't recommend these wholeheartedly.

15:10 And then Microsoft Copilot added a council and critique

15:13 feature which I thought was worth a quick mention.

15:15 Basically the council is something we already saw in Perplexity's product.

15:19 Talked about this in a previous episode where it consults multiple

15:22 models and takes the best of those and presents it to you.

15:25 This is an advanced workflow that some people are successfully

15:27 using for various things including investment advice in some cases.

15:31 But at the end of the day it's not as simple

15:32 as just asking multiple models and throwing it in somewhere and saying,

15:35 "Hey, summarize this." For example,

15:37 Microsoft Copilot takes an approach where it's two AI partners.

15:40 One is optimized for deep exploration and the second

15:43 one is there to validate all the claims,

15:45 improving the presentation and strengthening the structure.

15:48 Creates a feedback loop that is run through many times.

15:51 It's kind of a cool product that I would expect

15:52 we'll find in all of these AI super apps eventually.

15:55 And that's really it.

15:55 Another week on our journey to AGI, AI super app, whatever you want to call it.

16:00 That's what we're moving towards and this show is designed to keep you

16:02 up to date and to keep you at the frontier of all of that.

16:05 And as per usual, you don't have to be using all of these things,

16:08 but knowledge is power and so is awareness.

16:10 And now you're aware of all of this stuff.

16:12 And even though you might be spending your days in ChatGPT,

16:14 that's completely fine.

16:16 As per usual, I would push back heavily on this notion of, "Oh my god,

16:20 you're falling behind so heavily in AI just

16:22 because you're not experimenting with Open Claw and oh,

16:25 you're not building your own command center in Claude code.

16:28 What are you even doing?" No, that sells well.

16:29 That makes a good YouTube video.

16:31 But even some of these pioneers in terms of AI usage,

16:34 even if they use their Open Claws,

16:36 most of the time they use them similarly to how people use ChatGPT.

16:38 They ask questions, manage context,

16:40 and then they keep working with those results.

16:42 My hope is that between using this stuff regularly and you following this show,

16:45 you'll be able to stay at the frontier,

16:47 which hopefully should allow you to focus on the things

16:49 that you do best and the people that you love the most.

16:51 That's the true mission of why we spend time together here every week.

16:54 And it's my pleasure to be your guide on this journey.

16:56 All right, I'll see you next week.

16:58 AI news you can use.

17:01 Uh-huh.

17:00 One stop, one spot for the clues.

17:04 Hey.

17:03 Every week new keys, new views.

17:06 It's AI news you can use.

17:07 Let's go.