Could using AI to help you with your next big idea ultimately lead AI giants to beat you to the punch?
A professor and OpenAI are trading jabs over the progress made on a math problem so difficult to solve that there’s a $1 million prize for doing it. The entire saga raises questions about credit for AI-powered work and the data used to train models, writes BI’s Ben Shimkus.
NYU math professor Tristan Buckmaster announced major progress related to the Navier-Stokes problem only to be followed less than 24 hours later by OpenAI, which proposed a full-blown solution.
(Even if you skipped algebra in high school, stay with me here. I’ll keep it high level.)
Buckmaster was first to post about advancements he made tackling the problem with the help of AI models, including OpenAI’s. His post also detailed a tense back-and-forth with OpenAI about their competing research.
According to Buckmaster, OpenAI suggested coordinating any announcements and not mentioning Buckmaster’s partner, Levent Alpöge, who works for (wait for it) Anthropic. (You can read Buckmaster’s whole post here.)
OpenAI computer scientist Sebastien Bubeck pushed back against parts of Buckmaster’s post, saying he “never ever asked Levent to be removed from authorship of his own work.”
Buckmaster said OpenAI acknowledged working on the problem after it learned of his research, but told him “the model did not look up user data.”
On Tuesday, OpenAI published what it says is a solution to the Navier-Stokes problem. The company said it didn’t see Buckmaster and Alpöge’s work before they publicly released it, but also couldn’t rule out that “de-identified data derived from their usage of our products helped improve our models.”
You don’t have to understand long division to see some potential issues here.
Mainly, does using AI models mean you risk their owners front-running your idea?
According to OpenAI, its proof is very different from Buckmaster and Alpöge’s work. And Buckmaster didn’t accuse OpenAI of anything specifically, instead saying he just wanted to lay out the timeline of events.
But let’s say you had a killer idea for bridesmaid dresses you’re toying with in ChatGPT. Right before you’re ready to launch, OpenAI announces it’s releasing its own app focused on bridesmaid dresses. Coincidence … or conspiracy?
(Complex math problems are a lot closer to OpenAI’s wheelhouse than bridesmaids’ dresses, but you get the idea.)
This issue isn’t limited to academia and hypotheticals. Apple’s trade secrets lawsuit against OpenAI raised similar concerns. When a company’s secret sauce is shared with AI, it “may create irreversible and continually propagating uses of the trade secret.”
OpenAI has also discussed ways to benefit from users’ ideas that turn into big businesses thanks, in part, to its tech. CFO Sarah Friar pitched OpenAI getting paid when customers’ AI-enabled work makes money. Friar used a pharma partner as an example: When OpenAI tech helps develop a breakthrough medicine, it can take a licensed portion of the drug’s sales.
If everyone’s on the same page, that could be a great symbiotic partnership. But I’ll bet most of us aren’t well-versed in the terms and conditions of all the AI apps we’re dumping our ideas into.
