SWE's hot take: frontier models' raw intelligence hasn't improved nearly as much as hype suggests
Reasonable_Goat · reddit · 2026-11-15
A Reddit engineer argues LLMs got far more useful but their "raw intelligence" improved less than headlines suggest. Three observations: upgrading frontier models rarely improved a tuned production harness (failures trace to context engineering and tools); the same model varies widely across harnesses in complex codebases; capable local models suggest efficient problem-solving needs fewer parameters than assumed. He attributes much of the last 12-24 months' progress to cumulative model gains plus better harnesses, and suspects hype and doom narratives are co-fueled by upcoming AI IPOs.
More from AGI Musings
- A barber only heard of ChatGPT last week: real-world AI adoption gap — AI_Andrew · 2026-11-15
- Researchers push back on Dwarkesh: defining AGI metrics is the hard part — dhadfieldmenell · 2026-11-15
- AI researcher: nobody can tell X replies are bot-written — and that's scary — iScienceLuvr · 2026-11-15
- Pedro Domingos mocks AGI hype: AI researchers never feel AGI is near — pmddomingos · 2026-11-15
- Pedro Domingos asks: how much false info is in your brain from trusting LLMs? — pmddomingos · 2026-11-15
- Blogger signs petition to reinstate Jensen Huang, citing his open-weight advocacy — esrtweet · 2026-11-15