AI Researcher: Even 8 Hours of Agent Work Feels Brittle and 'Autocomplete-Like'
lateinteraction · x · 2026-09-20
Notable AI researcher lateinteraction posted a long thread checkpointing the current moment: models are still not good enough at real work. If you care about the shape of the work in more than 1-2 trivial dimensions, you'll quickly lose your mind over how brittle and autocomplete-like even 8 continuous hours of Astra or Fable agent work feels.
- Sparse cases of unimodal, verifiable success will keep multiplying, but on their own they tell us almost nothing.
- His core claim: something fundamental has not budged across model generations and 10,000x capability levels.
- He stresses he has no alternatives to sell and isn't discounting the progress or value created — if this changes tomorrow, he'll be the happiest person.
Related event: Researcher's rant on brittle frontier models sparks resonance(5 posts)→
More from AGI Musings
- ML author Burkov slams resigning AI scientists: 'What the hell was your goal?' — burkov · 2026-09-20
- Pedro Domingos: AI makes knowing what to do matter more than knowing how — pmddomingos · 2026-09-20
- SF's AI safety crowd sounds condescending — and it's fueling public resentment — signulll · 2026-09-20
- Vitalik defends EA ties while reasoning his way out of longtermism — inductionheads · 2026-09-20
- Noam Brown says air-gapping may not stop misaligned AI; Bryan Cantrill pushes back on "contagion of fear" — brianmichel · 2026-09-20
- Grady Booch amplifies bubble warning: the $500B AI infra mega-deal is mostly MOUs — Grady_Booch · 2026-09-20