Epoch AI researcher: most 'algorithmic progress' in frontier AI may come from better data
S_OhEigeartaigh · x · 2026-11-10
S Ó hÉigeartaigh highlights a thread by miclchen arguing that much discussion of AI R&D automation and recursive self-improvement rests on an outdated notion of "algorithmic progress." Citing Epoch AI's Anson Ho (Feb 2026), the thread claims most of what gets called algorithmic progress may actually stem from better data — a reframing with big implications for where the RSI feedback loop's real bottleneck lies.
More from AGI Musings
- '99% of people cannot extrapolate': AI revolution's lesson on imagining progress — daniel_mac8 · 2026-11-10
- A/B testing climbs the hill you're on — first ask if it's the right one, says Ronny Kohavi — richardreis · 2026-11-10
- iamtrask debates Davidson: labor-replacement feedback loop is scaling, not algorithmic RSI — iamtrask · 2026-11-10
- Gary Marcus: OpenAI and Anthropic token usage slipped over the past two weeks — GaryMarcus · 2026-11-10
- Best defense against full AI psychosis: always keep a wall-clock, data-constrained side project — shakoistsLog · 2026-11-10
- A personal benchmark system to track frontier AI progress and cheap-model ceilings — bendee983 · 2026-11-10