Post

Mark Dingemanse
@markdingemanse.net

When an "AI" provider announces a new result, remember they have a product to sell. Watch for the careful and systematic ways human labour is elided at every step and claims about the model and the 'discovery' process are shielded from direct inspection Today's specimen: openai.com/index/model-...

openai.comAn OpenAI model has disproved a central conjecture in discrete geometryAn OpenAI model solved the 80-year-old unit distance problem, disproving a major conjecture in discrete geometry and marking a milestone in AI-driven mathematics.
Mark Dingemanse
@markdingemanse.net

A single result presented, no account of how many attempts External experts paraded, no disclosure of internal checking Two paper-shaped objects, no notes on authorship, contributions, quality control A 'rewritten chain of thought', no access to the original and no disclosure of how, why, by whom

12 reposts56 likes

Nafnlaus 🇮🇸🇪🇺🇺🇦
@nafnlaus.bsky.social

You cannot brute force proofs like this. Which is why it hasn't been done before. Automated proof engines don't have a very dramatic track record. The most prominent one solved was the four color theorem, but it's proof is so long it can't be verified. Nothing at all like the present situation.

Mark Dingemanse
@markdingemanse.net

Two human authors worth reading on this: Lucy Suchman, Demystifying the intelligent machine: link.springer.com/chapter/10.1... Sarah Dillon, The Eliza effect and its dangers www.tandfonline.com/doi/full/10....

link.springer.comDemystifying the Intelligent MachineTaking inspiration from critical studies in the history, culture, and politics of technology, this chapter considers some contemporary developments in the project of creating humanlike machines. My ma...