Melanie Mitchell calls Stochastic Parrot critiques strawmen; lu_sichu pushes back
burny_tech · x · 2026-09-27
- Melanie Mitchell argued critiques of the Stochastic Parrots paper are strawmen: the paper addressed 2021-era LLMs, not today's AI systems with extensive post-training and external software components.
- lusichu countered that the paper's claims (§5, §6.1) were in-principle arguments about systems trained on form—fn. 22 already notes embedding an LM in a larger system doesn't grant communicative intent. If the argument fails for modern systems, it must be because they're no longer form-trained, not because it's 2026—unless midtraining, RL and pipeline changes actually overturn it.
- The exchange continues the long-running debate over whether LLMs possess genuine understanding or communicative intent.
Related event: Melanie Mitchell Sparks Debate: Today's Systems Are Not LLMs(31 posts)→
More from AGI Musings
- Not Boring: the three arguments that AI isn't a bubble are exactly how bubbles form — Not Boring (Packy McCormick) · 2026-11-11
- Viral post: five years of 'maybe next year,' then 2026 arrived — hey_abusiddik · 2026-11-11
- Jeremy Howard: LLMs "cosplay understanding," like the man in Searle's Chinese room — Machine Learning Street Talk · 2026-11-11
- OpenAI board member Paul Christiano: an AI accident is your most likely cause of death — notkilleveryoneist · 2026-11-11
- 'Chat with your data' misses the point — people want a smart coworker — vboykis · 2026-11-11
- Ben Todd: AI task time horizons appear superexponential in AECI, not just revenue — ben_j_todd · 2026-11-11