Kambhampati: Frontier models' 'jagged intelligence' traces back to opaque training data mixes
rao2z · x · 2026-11-08
ASU professor Subbarao Kambhampati cites a widely-shared tweet by Rohan Pandey in his talks: AI labs publish details on architecture, optimization, objectives, scaling, and kernels — everything except training data — leaving academia repeatedly astonished about where the secret sauce lies. Kambhampati argues that coming to grips with the so-called "jagged intelligence" of frontier models is impossible without acknowledging the role of opaque training data mixes.
More from AGI Musings
- Anders Sandberg: the real challenge is understanding where and why progress levels off — anderssandberg · 2026-11-08
- ChrisGPT: We could automate most work in 10 years, yet fear-driven bans dominate — ChrisGPT · 2026-11-08
- Z is Diophantine in Q proof slashes bounds: degree drops from 2↑↑7 to 4 — anderssandberg · 2026-11-08
- "All the bad stuff in society is going to go away": Brad Jacobs' AI optimism mocked as INFINITE FUN SPACE — generativist · 2026-11-08
- Researcher warns AI being right is the bigger danger: we may lose the ability to tell truth from falsehood — ValerioCapraro · 2026-11-08
- METR Assesses Anthropic's AI R&D Acceleration and Its Risk Conclusions — Matthew Berman · 2026-11-08