Gael Varoquaux says frontier LLMs still lag on tabular machine learning
GaelVaroquaux · x · 2026-07-27
Gael Varoquaux pushes back on the idea that LLMs have meaningfully replaced classic machine learning on tabular workflows.
In the exchange, he says he cannot tell whether the other person is trolling, but the substantive claim is clear: frontier LLMs, in his view, have never come close to doing well on good old tabular ML tasks.
The thread is a compact reminder that “LLMs everywhere” still has real limits, especially on structured data problems where bespoke models may remain superior.
More from Models
- Opus 5 Confidently Hallucinates and Refuses to Admit Mistakes — yacineMTB · 2026-07-27
- Moonshot AI puts Kimi K3 on Hugging Face with a launch countdown — Unusual_Guidance2095 · 2026-07-27
- Fable 5.1 looks strong as the author says model releases are speeding up again — iruletheworldmo · 2026-07-27
- Kimi K3’s pricing is being described as not cheap — ainch · 2026-07-27
- A long thread speculates that proto-GPT-6 may be earnest, puzzle-driven, and socially naïve — repligate · 2026-07-27
- Grok 4.5 now runs inside Codex with only an X Premium login — dr_cintas · 2026-07-27