ChatGPT Just Changed Forever...

0:00 It's a big week for Open AI.

0:01 Not just that they released a brand new flagship model,

0:04 GPT 5.5, but also they have a brand new image model.

0:08 Create social media graphics from this thumbnail in three sizes,

0:11 Instagram square, Twitter banner, Pinterest pin.

0:14 And look at that.

0:15 Instagram square, Twitter banner, Pinterest pin, all from one prompt.

0:21 [gasps] So, there's really something for everybody here.

0:23 Plus, a few more interesting AI releases that you should keep

0:26 an eye on in this week's episode of AI news in news,

0:29 a format that I recently reworked to be simpler

0:32 and to focus on the things that I actually care about

0:35 rather than rounding up all the AI news stories

0:37 that you could put to work today as a non-technical user.

0:40 Let's start by looking at GPT 5.5 cuz this is the big one.

0:43 And let me be up front, people, including me,

0:46 are still figuring out if this is going to make a difference in their workflows,

0:50 in the ones that they already have.

0:51 Because this model is not optimized for chat.

0:56 Igor, a new chat GPT model that is not optimized for chat?

1:00 What does that mean?

1:01 It means that AI is shifting.

1:03 My first impression of GPT 5.5 is that it is it is different.

1:08 As I talked about in the upload earlier

1:10 this week showing the three different levels of AI,

1:12 which I'll be referring very often from here on out, AI for answers,

1:16 AI as a work partner, and then AI working for you.

1:19 That top level is the agentic level where AI is doing things for you.

1:23 And this model has been developed for exactly that.

1:26 It's made for long-running tasks that are complex.

1:30 I mean, I sight.

1:31 It's designed for complex work like writing code, researching online,

1:35 analyzing information, creating documents and spreadsheets,

1:38 and moving across tools.

1:41 Now, sure, this happens in chats, too, but primarily these are agentic tasks.

1:45 Agentic coding is its headline capability,

1:48 and they keep reiterating this ability to take super messy

1:51 inputs and turn them into structured outputs that actually make sense.

1:54 Meaning, you can just bring a bunch of notes and say, "Hey,

1:57 turn this into a presentation." And it will do it for you.

1:59 Now, I'll show you some examples in a second here,

2:00 but across the internet, people are unsure.

2:03 Just like with Opus 4.7, the reviews are mixed,

2:06 and I think the reason for that is we've

2:08 sort of plateaued in terms of user-friendliness in chat applications.

2:13 And when it comes to writing code,

2:14 the bar is so high already with such good models out there that really

2:18 a lot of these new models just feel like a different flavor of really good.

2:23 But rather than just reciting what people say across the internet,

2:26 let me actually show you some examples of this in action.

2:29 And as always, I'll stay as transparent as possible.

2:31 This was the very first prompt we ran.

2:33 It's something that worked in Opus 4.7

2:36 when that came out last week, immediately.

2:38 And Opus 4.7 nailed it.

2:40 GPT 5.5 thought for 3 minutes and 12 seconds,

2:43 and it created this YouTube repurposing application,

2:47 which actually ended up not working on its first shot, whereas in Opus it did.

2:52 Oh, what a loser.

2:54 How do you access it in chat?

2:55 Well, that's actually quite simple.

2:57 You just go here to your model selector at the top,

2:59 and by default you'll probably be on instant,

3:01 and this model now hides behind thinking and pro on paid plans,

3:05 plus pro, business, and enterprise.

3:07 And you can see this clearly if you go to this configure option,

3:10 and you'll see that instant runs on the older model 5.3,

3:13 quick replies, and the new thinking and pro models,

3:16 if you pay for the $200 pro plan, are available in here.

3:19 So, if I switch to thinking, I can actually pick the thinking effort.

3:22 For the testing examples, we defaulted to extended because also Opus is not like

3:26 a heavy thinking model that goes ahead and thinks for 5 minutes,

3:28 so we tried to pick similar settings here.

3:30 Now, here's an interesting one.

3:31 I always run this prompt, right?

3:33 Create an SVG of the Death Star in the sky above Los Angeles,

3:36 just to show its ability to, I don't know, visually represent something.

3:40 And this is the first model that is actually, hmm,

3:43 I don't know, maybe too smart for its own good.

3:45 Look, every time I try this, it starts thinking through this process,

3:47 and it's like, "Hmm, do we really want to make an SVG?

3:50 Maybe I should make an SVG manually." Or just make it.

3:53 Come on.

3:54 In this case, it gave me a path to file I can't even access.

3:57 In a previous testing, it sometimes switched to the image tool,

4:00 and it generated a PNG of what I asked for.

4:03 And then when re-prompted, it got this data path again.

4:05 So, not perfect.

4:06 It's sort of like overthinking it.

4:10 Again, it's the point that I started with.

4:12 I don't think this model is made for chat, and that's fine.

4:16 I will be testing this with my agentic setup.

4:18 I want to see if this performs with my open cloth.

4:20 But in chat, I don't know.

4:21 I don't think I've seen a release where out

4:23 of the few basic test cases that we run,

4:25 as I just show you, multiple don't actually work.

4:29 Now, I want to be clear, this is not me saying that this is a bad model.

4:32 For example, we ran another test prompt

4:34 where it builds this, um, space shooter game.

4:37 And this is by far the best space shooter I've ever seen out of an LLM.

4:41 Look at the complexity of this thing.

4:43 This actually looks great, 3D space.

4:46 Look at all that.

4:47 So, clearly it's doing something right.

4:49 But what I am saying is that I don't think this is a model made for chat.

4:52 And my conclusion on this is hmm,

4:56 I think Open AI is getting ahead of its current state of chat GPT.

5:01 And what I mean by that is that in the next week or two,

5:04 or I don't know, maybe three or four, but definitely shortly,

5:07 I think we'll see an overhaul of chat GPT and it will turn into something

5:11 that is a bit more agentic because

5:13 clearly that's where the entire space is moving.

5:16 And they are the prime consumer application,

5:18 and maybe they're just putting this model in there now as a first step.

5:21 If you're making stuff, it's great, you're going to love it.

5:23 For everyday tasks, it's similar.

5:25 But for some of these specialty tasks,

5:26 it just seems like chat is a bit too limiting for what it actually wants to do.

5:30 Okay, so that's GPT 5.5.

5:32 But if you're using a chat GPT,

5:34 there's another big release that happened this week,

5:36 and this one is actually exciting.

5:38 So, you know they had an image generator.

5:40 Now, there's a new image generator.

5:41 It's called GPT Image Tool, and it's so fun.

5:44 First and foremost, what I tested right away

5:46 is how good is it at retaining people's faces.

5:48 This was a big thing inside of Nano

5:50 Banana and why it was so popular with people.

5:52 You can upload an image of yourself, like me right here, and say something like,

5:56 "Create professional headshots of me." And then look at these.

5:59 All of them actually maintained my face.

6:01 They didn't try to regenerate another person who kind of looks like me.

6:05 This is actually me.

6:06 Big win.

6:07 And this is accessible to all chat GPT users, including free accounts.

6:11 Really great stuff.

6:12 Now, a few more things to be noted here.

6:14 First of all, I believe this is the first imaging model from one

6:17 of the big providers that allows you to turn on thinking mode.

6:21 So, in practice, if you enable image creation like so,

6:24 you could also go up here and say thinking, and all of a sudden it will think

6:29 multiple rounds before it goes ahead and creates image,

6:32 which can be really good with, well,

6:34 something like animal infographics right here.

6:36 Let's just go with the preset example.

6:38 And while that generates, let me show you what we actually did with this cuz I

6:42 thought this one was really impressive because it's not just good at creating,

6:45 but also editing and repurposing.

6:48 One of the big new features now is it can output up to eight different images.

6:51 So, I just went ahead and uploaded a recent YouTube thumbnail.

6:54 This was the video on the free level framework.

6:56 I highly recommend watching that if you're a long-term viewer of the channel.

6:59 I'll be referencing this over and over again.

7:01 It's really good.

7:01 It talks about how to get into agents.

7:03 And I said, "Create social media graphics from this thumbnail in three sizes,

7:06 Instagram square, Twitter banner, Pinterest pin." And look at that.

7:10 Instagram square, Twitter banner, Pinterest pin, all from one prompt.

7:16 I am critical of these things.

7:18 The composition is good.

7:20 No, it's not good, it's perfect.

7:22 Look at it resizing the font in a way that is actually tasteful.

7:26 To me, this is unbelievable.

7:28 It does the work of a graphic designer, right?

7:30 You could now repurpose one thing into many others easily.

7:34 Beyond that, it has near-perfect text rendering.

7:36 It can do entire restaurant menus,

7:38 and it's not like every fifth word has a little mistake.

7:40 Now, it's actually near-perfect.

7:42 It can output 2K resolution.

7:44 It various formats as you just saw right there.

7:46 And they dramatically improved the instruction following,

7:49 meaning if you tell it to include certain words,

7:51 it will actually include them rather than take creative freedom with it.

7:54 And yeah, also as I showed you,

7:56 the ability to actually retain your face and like

7:58 remix yourself into different situations or create professional headshots,

8:01 that's a really popular use case, is,

8:03 I would even argue, better than Gemini's Nano Banana Pro.

8:06 Overall, I'd state that this is probably the best and certainly

8:09 the most user-friendly and useful image generator out there now,

8:13 and it's available on all accounts.

8:15 Go and try it now if you haven't done so yet.

8:17 And those are the main stories for this week,

8:18 but there are a few things that popped up on my radar that I wanted to share.

8:21 Again, as I shared last week,

8:22 I want to drop this responsibility of covering everything

8:25 that came out in AI that you could use.

8:28 I'm rather I'll do this radar segment

8:29 of like interesting stuff to keep an eye on.

8:31 And one of them is Claude Cole work adding dynamic artifacts.

8:36 And listen to this, if you code up an application,

8:38 it can become a permanent part of your workspace.

8:41 And then you can pin it to the side, and application can now,

8:45 and this is the kicker, take in live data from your connectors.

8:49 So, if you use some application that works

8:51 with your existing emails and you connect it to Gmail, well,

8:54 every time you open the app,

8:55 it will pull in your newest emails or any other data from another connector.

9:00 This makes creating custom apps with custom

9:03 data so much simpler than ever before.

9:05 It's so fascinating to watch

9:06 these user-friendly personal agent type applications develop.

9:10 And as I said, I think Open AI will be following suit

9:13 here because they'll need a Cole work type application in their own ecosystem.

9:16 And I think they're going to evolve chat GPT to that very soon.

9:20 Also interestingly, and this is more of the developer side of things,

9:23 if you're building something or want to use this, there's

9:25 a deep research Max API now from Google Gemini.

9:29 So, if you're familiar, it's basically deep research,

9:31 but it can connect to your data,

9:33 the web, and generate full reports in just one API call,

9:36 which just crushes all benchmarks.

9:38 This seems like the best research tool on the market.

9:41 And then also, finally, I want to mention Claude design that came out last week,

9:45 which was Anthropic's first stab at a design tool.

9:48 Looks really solid, and people are using it.

9:49 You create marketing materials, pitch decks, app prototypes,

9:53 and you could even export directly to Canva or PDF.

9:56 And this makes them the first major AI company

9:59 to ship a full visual design product inside the chatbot interface.

10:03 Worth looking into if you make design assets

10:05 or are even into something like email marketing,

10:08 social media content creation, or similar.

10:10 Heck, people are even experimenting and making

10:12 these little PowerPoint style videos with it.

10:14 And that's everything I wanted to share with you today.

10:16 Busy week.

10:16 We've been preparing and holding the AI Advantage Summit,

10:19 online conference designed to save you time with AI.

10:22 And I just want to again highlight that this is

10:24 really geared at people who are not at level three yet.

10:26 We're not building agents.

10:28 Entrepreneurs, small business owners,

10:29 or professionals who are non-technical and are

10:32 figuring all of this AI stuff out.

10:33 If you're interested in that, it was also re-streamed on this channel.

10:36 You can find it under the live recordings for the next 7 days.

10:38 And with that being said, my name is Igor, and I hope you have a wonderful week.