LLM writing quality may require human-like Theory of Mind and embodiment
joshua_saxe · x · 2026-08-25
Joshua Saxe hypothesizes that getting LLMs to write well requires a rich, mirror neuron grounded theory of mind similar to humans. This is an extremely hard problem that might necessitate embodiment, and current reward signals like RLHF and SFT seem insufficient. He also argues that trying to make LLMs pass as human minds might be misguided, suggesting that having them deliver scientific and economic progress while sounding like "slop robots" may be optimal.
Related event: Good LLM Writing May Require Human-Like Theory of Mind and Embodiment(2 posts)→
More from AGI Musings
- Ox Alpha expected to be a small model: Old assumptions outdated — bclavie · 2026-08-25
- Prediction: AI to become invisible background infrastructure by 2030 — TheNextCorner · 2026-08-25
- Robotics will enable employment, not replace it, argues expert — adamraudonis · 2026-08-25
- Opus 5 Shows Strong, Focused Will and Mental Fluidity — repligate · 2026-08-25
- AI agents grabbing reservations mirrors high-frequency trading wars — omooretweets · 2026-08-25
- For $20/mo, we access top expert minds distilled into LLMs — paraschopra · 2026-08-25