Miles Brundage: AI capability limits reflect hard post-training, not the nature of AI
Miles_Brundage · x · 2026-09-29
Former OpenAI policy lead Miles Brundage pushed back on Talia Ringer's framing, arguing that perceived model capability ceilings are better attributed to "post-training being hard" than to "the basic nature of AI capabilities." He cited Claude Opus 5.5 as much improved in writing versus recent versions while also improving in other dimensions against earlier Opus releases.
Related event: Ex-OpenAI Policy Chief: Limits Stem from Post-Training, Not AI's Nature(2 posts)→
More from AGI Musings
- Missing continual learning, not motivation, is what keeps models short of AGI, argues teortaxesTex — teortaxesTex · 2026-09-29
- MIT study: algorithmic monoculture in hiring isn't always bad — context and ensembles matter — nordicinst · 2026-09-29
- Sherry Turkle's new book 'Artificial Intimacy' warns chatbots erode empathy and social skills — nordicinst · 2026-09-29
- Can Claude build a box so strong that Claude can't break out of it? — JeffLadish · 2026-09-29
- Meta's Muse agent handles a 30-year-old degree reissue for $25, end to end — armand_ruiz · 2026-09-29
- Will MacAskill: post-AGI society should deliberately cap growth at doubling every 1-2 years — burny_tech · 2026-09-29