(Preview) The Delights of Images in GPT, The Future of Graphic Design, Signal and Multiple Dimensions of Security

Reactions to OpenAI's "Images in GPT," and thoughts on the Trump administration's Signal debacle, including a few points Ben neglected to emphasize earlier this week.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Andrew Sharp?

  Hello and welcome to a free preview of Sharp Tech.

• 0:08Andrew Sharp?

  What is interesting to me, though, is I feel like we've, should we presume it's only a matter of time before everyone else in consumer AI can offer this functionality? Like, the example from Google, it came earlier, it looked worse than what I was seeing from OpenAI, but they're clearly in the same ballpark already. Is it only a matter of time before we can all generate horribly explicit images with Grok?

• 0:34Ben Thompson?

  Yeah, we need Grok to comply again. I think they do have something along this line in the pipeline. Yeah, I mean, this, obviously, this is going, this is going to be everywhere. I mean, and it was pretty interesting that, you know, OpenAI was pretty explicit that we're taking the brakes off a little bit. You can, like, it's interesting because the number one objection, you hear this from, from people in DC, is politicians are just terrified of, like, fakes, of, like, being in there. And so there, and so that's always been, you know, and by and large, the AI companies have adopted a, okay, just don't put celebrities in there at all. OpenAI in this case was like, okay, now it's an opt-out.

• 1:17Ben Thompson?

  You can render celebrities, but if the celebrity or whoever can reach out and say, don't allow me to be rendered, and then that will sort of go into the system. And, which I, you know, I'm sure some people will be, will be all over. It does accelerate, though, just this very rapid, ongoing, arguably been the case even before AI, but AI is making it a gazillion times worse, sense that everything on the internet is fake. And the ones that resonate, what resonates has some connection to reality, but the actual output is not reality, and that fake content is just multiplying very rapidly.

• 1:58Andrew Sharp?

  Well, and it's funny, I think that is why there was more virality with this new update, because, like, I enjoyed seeing your cat stylized. I enjoyed seeing people post- Photos of their kids stylized or a guy with his wife on a boardwalk like that's more interesting to me and some of the Dali images, including some stuff that I saw on Stratechery. It felt like we had sort of plateaued in terms of what was possible there. And it was all kind of a steampunk aesthetic that didn't really resonate with me. And so personalizing it definitely makes a difference in terms of striking a chord with the mainstream.

• 2:40Ben Thompson?

  Yeah, there's there's something about this connection to something real that I think resonates with people more than just sort of pure ethereal generating something and. You know, which I again, I'm still wrapping my mind around this. I feel like there's there are some takeaways about why this happened to go broadly generally. That we should continue to unpack, but um, but maybe, but alas, I was thinking too much about other.

• 3:11Andrew Sharp?

  Signal. Well, I mean, the other thing that you wrote on Wednesday was the images in chat GPT actually makes real a lot of the fears that accompanied the first wave of image generators. So essentially by having the ability to craft images in more detail or alter images in more detail. That's where the concerns around deep fakes become more urgent. Is that right?

• 3:39Ben Thompson?

  Well, I was thinking more of the graphic designer angle that I was talking about.

• 3:42Andrew Sharp?

  Okay.

• 3:42Ben Thompson?

  You know, just a couple of minutes ago where like they're you could theoretically see image generation is bad for graphic designers was like, well. Part of being a graph designer is actually being precise and getting the information right. Obviously not just text being right, but the positioning and the balance and all the things that go into graph design. This is a workflow like it's not just a one shot pray cross your finger. Hope it comes out like you wanted to. It's a, okay, here's a starting piece. Now we're going to iterate this. We're going to do this. We're going to combine this, put this together there. It's it and the emergence of a workflow is is.

• 4:19Ben Thompson?

  Arguably, the more concerning part. Again, it's kind of a dumb point to index on because once you got to image generation, it was inevitable we were probably going to get here. But it does feel like a step change to me in how much control you have.

• 4:34Andrew Sharp?

  Totally. Well, and control is what's required in order to deploy these tools in a professional setting and rely on them. And it's similar to like Sam Altman, you interviewed him about this a little bit, but he came out with the OpenAI fiction generator, which is kind of interesting. And I'm curious to see what emerges on that front over the next couple of years. And the idea of hallucination being a good thing is kind of fun. But like in a work setting, you want something like deep research and deep research is what makes you sort of sit up in your chair.

• 5:10Ben Thompson?

  You still worry about hallucinations, to be clear.

• 5:13Andrew Sharp?

  To be clear, we should add that disclaimer every time we mention deep research. But you know what I mean? Where it's like, the ability for AI to play God and create something whole cloth is cool. But in terms of practical impact, at least in the short term, the ability to just do exactly what you want and reproduce a good idea in eight different ways and let you choose, like, that's a real game changer.

• 5:39Ben Thompson?

  No, you're helping me collect my thoughts on this. I think you identified a couple of things. So we already established a connection to reality resonates more. What you just put your finger on is the sense of feeling in control, I think, is something that is important and resonates. And number three, I would say, is this sense of

• 6:08Ben Thompson?

  needing a prompt is an obstacle. Like, a blank page with a cursor has been the fodder for nightmares for students for ages and ages and ages, right? Like, you know, what I've dealt with graphic designers in the past or deal with people that are doing things, it's always the, just generate something and that will prompt me. To figure out what my actual thoughts are. And oh, no, this should be there. This should be that. If you just want me to give you my opinion from scratch, it's actually very, very difficult. And it's interesting. I wonder if there's an aspect of the design of these tools is mistaken in that they're predicated on a user prompt. And maybe the real breakthrough

• 6:51Ben Thompson?

  is the system prompt. Now, the system prompt in the context of AI is what's hidden in there. It gives the AI all the instructions of what to operate. When I say system prompt in this context, I mean, the prompt they give the user. How do you trigger the user to do this sort of thing in a way that's not annoying and whatever it might be? You go back to those viral apps, it was pretty straightforward. Put a headshot in, right? And there's a constraining of what you can do that actually meets far more people where they are. And this is part of becoming a product company. It's part of like, not just having an API that you can do whatever you want with. It's actually making choices, constraining your product,

• 7:32Ben Thompson?

  and making it clear to someone what they can do. Now, in this case, OpenAI didn't, you know, as far as I know, it wasn't some OpenAI AstroTurf campaign to get people doing Studio Ghibli. This was a viral moment. But it was easily, easily enough achieved that it felt like, oh, yeah, get a photo, put it in your chat app. And people know how to get photos and put it in a chat app. And then it took off. But that is, it's just, I don't know, there's lots of, like I said, I just think there's lots of really weird, interesting little lessons with this entire episode.

• 8:08Andrew Sharp?

  Okay. Well, I look forward to continuing all of this conversation in the weeks to come. Do you want to hit the signal thing with the second half of the episode here?

• 8:17Ben Thompson?

  I do. Look, like I said, I've put a bit of a mea culpa at the top. When this whole signal thing drops, so let's, let's just sort of zoom out, big picture. And you're the DC guy, so feel free to, feel free to jump in.

• 8:33Andrew Sharp?

  All right, and that is the end of the free preview. If you'd like to hear more from Ben and I, there are links to subscribe in the show notes, or you can also go to sharptech.fm. Either option will get you access to a personalized feed that has all the shows we do every week, plus lots more great content from Stratechery and the Stratechery Plus bundle. Check it out, and if you've got feedback, please email us at [email protected].