AI Signal 513 2 feeds carried it
OpenAI cannot rule out user chat data contributed to its math breakthroughs
↗
A second mathematician has accused OpenAI of dishonesty after the company declined to conclusively rule out that interactions users had with ChatGPT entered its training data and contributed to its mathematical results.
For anyone building on or with OpenAI's models, this raises the question of whether proprietary or unpublished work shared through chat interactions could be absorbed into training data and reproduced without credit. OpenAI's position, that it cannot rule out indirect influence from de-identified user data, means there is no guarantee of confidentiality in model interactions.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Mathematician Andreas Thom accused OpenAI of dishonesty after the company's response about training data did not rule out that his ChatGPT interactions contributed to its non-sofic groups result.
OpenAI acknowledged its result built on previous work by Thom and Gábor Kun and quietly amended its writeup, but stated it 'cannot rule out that de-identified data derived from their usage of our products helped improve our models.'
Thom argues that only OpenAI can prove whether user data was used, since researchers cannot reverse-engineer its training pipeline.
THE CLUSTER