Accelerating science with Prism

Ever since we launched GPT5, we're seeing increasing evidence that AI can accelerate science. Just in the last few weeks, we've seen a number of examples where GPT 5. 2 has contributed solutions to open mathematical problems. And we're seeing the same across biology, physics, chemistry, material science, and more. When you look at scientific tooling, a lot of it hasn't changed in decades. So part of the motivation for Prism was what if we could give every scientist AI superpowers. Now part of that of course will be about pushing the frontier with our models. But it isn't just about the models. We also want to make it easier for scientists to use AI by bringing it to them where they work. That's why we're so excited to introduce Prism. Prism is a free AI native environment for scientific writing and collaboration. And if we're successful, it means less time spent in your editor and more time doing research. — Students remember like, okay, actually don't hit that marker. — Prism started with the observation that the real unlock as a developer was having the AI directly in your editor environment. So you didn't have to go back and forth between, you know, chat and your coding editor. It was just there as a partner thinking with you. And we really noticed that with scientific workflows, they didn't have this kind of back and forth conversational aspect that we saw as developers. So now that you know a little bit about why we built Prism, I think instead of talking about what Prism does and how it works, maybe you could just show us, Alex. — Yeah, let's do it. — Yeah, absolutely. So let me just say um I've been writing physics papers now for my whole career and I'm really excited that Prism is finally here. I've been using Chad GPT in parallel to my latte editor going back and forth and now I don't have to do that anymore because I can call Chad GPT directly within the Latte environment and work on my papers directly. So here's a draft of a paper I wrote recently. But at this point, although my writing is very good, I think Chad GPT writes even better. And um just improving the writing um is such a great use of the tool. So I'll say periphery the abstract going line by line. I'm very particular. So I want to control everything that it says. And so now you have it's like you had uploaded your entire project to chatgbt and we're asking the same thing except here in prism automatically your whole project is in context. — Exactly. So here it went through uh the text that I had written which is now shown in red and it's suggesting these edits in green. I can check um what exactly it's suggesting to be changed. It's highlighted and you know in this case I approve of the changes so I can say keep. But this is just scratching the surface. Another thing that is so annoying about latte is producing beautiful technical diagrams which can take a huge amount of work. — Oh my god. — Yeah. — However, with Prism now, we can ask Chad GPT to do this painful work. So, here's a picture of a commutative diagram that I drew on a whiteboard earlier. And let's see if we can get Chad to turn that into a perfect figure for my paper. — Yeah. While he's typing the prompt, you can draw a diagram here in 20 seconds. And then, ironically, when you go to bring it on to your computer, it takes you an hour to figure out exactly how to get it to work in Latte. — If you're lucky. — Yeah, if you're lucky. It's like a bunch of searching on the internet and moving things up and down, pixel by pixel. — Render, rerender. — So, I mean, I wish I had Prism in grad school. Okay, here we go. So, I uploaded the picture of the diagram that I drew on the whiteboard, and I instructed Chad to turn this into a Tix diagram and insert it where my cursor is. That's where I want the figure to go. And I'm asking it to um be faithful to the geometry of the picture. — And let's see what it does with that. Ah, here we go. It produced a tix diagram. Let's accept that and compile. And here we go. Pretty good. Wow, that's really quite excellent. This is almost a perfect oneot for a really complicated diagram. Another feature that I love about Prism is that I can actually create multiple instances of chat to work on different aspects of the paper at the same time. So one extremely timeconuming task is to look up references figuring out everything in the literature that is relevant for your paper. And then grabbing those references and putting them into the latte source in the right format is extremely painful. Let's ask chat in this new little window that I can create by clicking this button to do that for me. So these new windows still have the full context of your entire project in them. So you don't have to uh upload anything. You can just reference the paper. — Yeah. So now I'm just asking chat find all the relevant papers in the literature that I should consider citing. This is exactly the kind of thing that I would normally do by copy pasting the latte source or — uploading the PDF into a new window on chat GPT and asking it to do the exact same thing. But now I don't have to explain what the context is because it knows I'm working on this project. And when I'm saying find relevant papers, it knows what I'm talking about. And it's going to take some time to do this. In the meantime, hey, um, there's a calculation here. I write down this generator of a symmetry of a

differential equation. Perhaps I'm not totally sure of the answer. Let's say I want to check it. I can create a new chat window and get it to do that while the other chats are doing their own thing. So I can launch many tasks at the same time. — You have a whole team working for you. — Yeah, absolutely. So now I'm going to tell it in the new symmetries section there is a differential operator H+. Um check that it generates a symmetry of the stationary and axis symmetric wave equation. And here you don't want it to check in the paper. you wanted to just look at the paper and then show you in the chat window. — Show me the calculation here. Okay. So, let's get it to do that. In the meantime, let's go back to the literature search. Ah, here we go. It came up with a beautiful list of papers that are all relevant to the paper. Wow, this is great. And each thing has its own different format and another thing where you just spend hours searching around on the internet to try and you know figure out exactly the right way to do things in a way that doesn't add to science at all. It's just kind of busy work that if the AI can do for you and can you know also check its work and ground it as it goes — you can do more of what you actually you know became a scientist to do. — Okay. Right. So while we were chatting here having a good time um just from this simple prompt it understood that I was talking about this generator it grabbed the equation from above uh I didn't have to copy paste the equation itself it already read it from the draft which is in context understood exactly what I was asking and now it's explaining the calculation that must be done to check the correctness and now it does the symmetry check what's great about this is that we're calling Chad GPT thinking mode. So, we're always getting to use the uh best version of the model that we can and it's now really capable of doing checks of the math in the paper. So, it's not just proofreading the text, but it's actually also engaging with the substance of the paper. I think um the existence of Prism and AI more generally is going to really accelerate science and uh remove a lot of the drudgery leaving more time for the truly important and uh joyful tasks. — Yeah, which is exactly the goal. We've talked a bunch about 2026 and how it's going to be a huge year for AI and science. you gave the example of uh 2025 and everything you learned about AI and software engineering and the power of bringing AI into the the actual workflow. And that's why we're so excited to put Prism into everybody's hands so that we can see that play out across the whole world. —