Why Anthropic's 2x capability gains won't translate to RSI: manual post-training and async autonomy
menhguin · x · 2026-11-11
menhguin introduces what he calls "slowness of RSI": even granting that Anthropic is likely improving models on well-specified tasks at 2x the rate of March 2024–October 2025 (as of October 2026), this won't escalate to recursive self-improvement. Key blockers: most tasks aren't well-specified, post-training pipelines remain manual, finding good autonomous/scientific research loops takes time, and async autonomy is unsolved — all arguments against very short timelines.
More from AGI Musings
- Viral post: five years of 'maybe next year,' then 2026 arrived — hey_abusiddik · 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
- When AI becomes an oracle, what happens to mathematics? — stevenstrogatz · 2026-11-11
- David Patterson: superintelligence will end profit and equity growth, only sales-tax-funded universal high income works — davidpattersonx · 2026-11-11
- Even after a 6-month recursive leap, AI progress stays 3x the 2024 rate — ben_j_todd · 2026-11-11