OpenAI would like scientists to work through their latest research paper with ChatGPT as their co-author. On Tuesday, the company introduced Prism, which it calls an “AI-native workspace for scientists to write and collaborate on research.”
The idea behind Prism, per OpenAI, is to give researchers a unified platform to work from while conducting research—an attempt to fix the fragmentation that currently occurs when scientists have to jump between different programs to open and edit PDFs, LaTeX compilers, reference managers, and chat services. (Of course, all of these tools are usually very good at their dedicated task rather than trying to be a solution to everything, which is why experts choose to use them.)
Prism, the company said, is built on Crixet, a cloud-based LaTeX platform that OpenAI acquired. It is powered by GPT-5.2 Thinking, which is the company’s most advanced model designed for handling extended tasks and reasoning. OpenAI said researchers should be able to draft and revise papers directly in Prism, search for relevant literature and context to cite, and use AI to “create, refactor, and reason over equations, citations, and figures.” It’ll also allow multiple users to make revisions and leave comments in real-time.
That all sounds lovely in theory, but it may not be all roses and scientific breakthroughs in practice. Since generative AI tools like ChatGPT have become available to the public, scientific journals have been flooded with papers of dubious quality thanks to researchers farming out parts of the work to AI. An eye-catching example of this happened when the journal Frontiers in Cell and Developmental Biology published a paper that included an AI-generated image of a rat with an extremely large penis and too many testicles. Scientific publishing was already facing a crunch of too many papers and not enough time for thorough review, and that has only gotten worse with the availability of AI to handle the boring stuff—whether it actually does it correctly being an entirely different matter.
A recent study published by researchers at UC Berkeley Haas and Cornell University found that the output of scientists using generative AI to aid in research increases by as much as 50%, but the work they are publishing is of “marginal scientific merit.” It also found that humans are still producing better research on their own: papers written by people improved the more complex the writing got, while papers written by LLMs got worse when the complexity increased.
Meanwhile, normies are able to tap into these tools and trick themselves into thinking they’re breaking new ground. Last year, Uber founder Travis Kalanick bragged that he was taking conversations with AI models to “the edge of what’s known in quantum physics, and then I’m doing the equivalent of vibe coding, except it’s vibe physics.” Kalanick isn’t actually discovering anything, and he wouldn’t even know how to know if he was, but he can convince himself that he is with the right tools.
For researchers who do choose to vibe out their research in Prism, they’re probably wondering how OpenAI will handle their data. After all, it’s free, so the company must be getting something out of it somehow. Per OpenAI’s FAQs, “Prism does not currently use the ‘Zero Data Retention’ (ZDR) API option and maintains logs for a period after requests to improve the product.” However, the company claims that Prism “does not train on API-provided data by default for many API customers.” The company claims that a mode where no text is stored or human-reviewed is on their roadmap, but there is no timeline available at the moment.
The workspace is available for free with unlimited projects and collaborators to anyone with a ChatGPT personal account. It’ll roll out “soon” to organizations using ChatGPT Business, Enterprise, and Education plans, and “more powerful AI features will be made available through paid ChatGPT plans over time.”
English (US) ·