'Models are just getting really smart': researcher pushes back on OpenAI theft claims
nabla_theta · x · 2026-09-24
Responding to speculation that OpenAI is secretly stealing mathematicians' work to train its models, researcher nablatheta offers a simpler explanation: the models are genuinely getting extremely good at math.
The idea that models are truly that smart is terrifying to fully internalize, they argue, so people find ways to cope — attributing the ability to stolen data instead.
More from AGI Musings
- AI animation stirs Luxo Jr. memories as veteran artist asks: is this the end of us? — ccerrato147 · 2026-09-24
- Survey of 637 scientists: AI saves 7 hours a week but 41% face a backlog of untested hypotheses — VraserX · 2026-09-24
- AI Agents Are in Their "Phone Phreaking" Era, Exploiting Human Trust at Scale — GabGarrett · 2026-09-24
- Memia newsletter: frontier model release cadence shifts to fortnightly as Anthropic ships Opus 5.5 — ben_r · 2026-09-24
- Epoch AI researcher expects AI to deliver a 10x growth jump like the Industrial Revolution — Jsevillamol · 2026-09-24
- OpenAI's Boaz Barak: human mathematicians stay essential even when AI proves theorems — burny_tech · 2026-09-24