Ex-OpenAI Employee Starts Company for High-Quality RL Datasets, Citing Poor LLM Generalization
rickasaurus · x · 2026-07-30
An departing OpenAI employee noted that despite billions invested, LLM generalization remains poor with "spiky" capabilities—such as coding models still requiring manual PR cleanups. He highlighted that public data is exhausted and hard to synthesize.
To solve this, he is launching a new company focused on producing high-quality reinforcement learning (RL) datasets. Industry experts echoed this, emphasizing that analytical and clickstream data is now extremely valuable as it cannot be generalized across companies, making it key to future model improvements.
Related event: Ex-OpenAI Staffer Leaves to Build High-Quality RL Data Firm(3 posts)→
More from Companies & People
- Inside Crosby: The 'Neo-Firm' Blending Lawyers and Engineers for AI-Era Legal Work — graceisford · 2026-07-31
- Tracking Leopold Aschenbrenner's Startup Drama: AI Reconstructs the Timeline — ivan_bezdomny · 2026-07-31
- Leaked Court Docs Reveal Anthropic Secretly Shredded Millions of Physical Books for AI Training — AICopyLab · 2026-07-31
- Seeking Advice: What to Expect in an OpenAI Security Engineer Coding Interview? — CircumspectCapybara · 2026-07-31
- Leopold Set to Marry Anthropic Chief of Staff, Shaking AI Safety Circles — beffjezos · 2026-07-31
- Giving Up the Model Race? Mistral Pivots to Become a European Palantir — BankZan · 2026-07-31