OpenAI Can't Rule Out Buckmaster Data Trained Models — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI stated it "cannot rule out" — though it considers it unlikely — that de-identified data derived from Buckmaster's and Alpöge's use of its products helped improve its models, according to the company's announcement.
- OpenAI denied its researchers or models saw Buckmaster or Alpöge's prompts directly, and disclosed it spent millions in compute after learning of an Anthropic breakthrough, per coverage cited by TechMeme.
Why it matters: The distinction between "saw the prompts" and "data derived from use fed training" is the substance of the dispute: OpenAI is conceding a privacy-relevant feedback loop while denying the more alarming direct-access claim, which matters for anyone whose AI usage could end up shaping future models.
Ask SkimNews
