Datology launches Curation Studio, ships DatBench VLM eval benchmark
josh_wills · x · 2026-10-09
Data curation startup Datology AI launched Curation Studio, a customizable frontier data curation platform for teams training their own models, premised on the idea that data quality is "the ultimate compute multiplier."
- Alongside it: DatBench (arXiv:2601.02316), a new VLM eval benchmark built on faithfulness, discriminability and efficiency — finding up to 70% blindly-solvable questions in some evals and up to 42% mislabeled/ambiguous samples in certain datasets
- Prior tech reports (DatSmith, BeyondWeb, UberWeb, pretrain-vs-CPT analysis) detail practical pre/post-training data curation tricks
- Commentators call Datology one of the best data research labs in the field
More from Companies & People
- 3 weeks in: 29.4% of the Fortune 500 showed up and a16z led a big Series A — airesearch12 · 2026-10-10
- From Bloomberg journalist to Anthropic cofounder: the Jack Clark career anecdote resurfaces — kevinakwok · 2026-10-10
- Three Lab Stories, Zero Ambiguity: Firing Safety Teams vs Banning Mean Talk to AI — paulnovosad · 2026-10-10
- Ex-colleague says Sam Altman aspired to be 'a great man of history' like Churchill or Napoleon — GarrisonLovely · 2026-10-10
- NYU mathematician says OpenAI's 722-paper AI math dump wrecked research careers — tedmitew · 2026-10-10
- Anthropic accused of calling RSP 'commitments' while dodging legal binding force — Miles_Brundage · 2026-10-10