MIT's Frankle: AGI-level intelligence already exists — the bottleneck is evals, not capability
jefrankle · x · 2026-10-22
MIT researcher Jared Frankle, after a conversation with Mercor co-founder Brendan Foody, agreed with Databricks CEO Ali Ghodsi: intelligence capable of AGI has existed for a while. The real bottleneck isn't intelligence but data context and verification — knowing whether AI did the right thing in your specific data context. The hard part is creating a measurable specification for qualitative tasks in real data, i.e., an eval, which is also the main obstacle to enterprise AI/AGI adoption.
Related event: Eval, not intelligence, is AI's real bottleneck: Frankle(3 posts)→
More from AGI Musings
- Audio and video are AGI's blind spot: models can't verify what they hear or see — coldrolledpotmetal · 2026-10-22
- If AI can reliably spot people who never publicly say racist things, what should happen to that info? — PierceLilholt · 2026-10-22
- New Paper Uses 288M Job Postings to Measure Where AI Bottlenecks Raise Work Value — soumitrashukla9 · 2026-10-22
- Aaron Levie: autonomous domain-specific agents are AI's emerging form factor, a leap past the LLM-wrapper era — _AustinCalvert_ · 2026-10-22
- Widening AI Ceiling-Floor Gap Means More 'Out of Distribution' Events — curious_vii · 2026-10-22
- Google researchers test an AI job-impact site: 'bluer the collar, the later the deprecation' — giffmana · 2026-10-22