Model Capability Distribution: Uneven Gains in Math and CS
scaling01 · x · 2026-08-27
The author argues that median metrics mask significant progress, noting that models are much more capable at math and computer science compared to other domains. The capability envelope is described as a highly non-convex 'shoggoth' rather than a simple median circle.
Related event: Frontier Models' Cyber Capabilities Lead General Ability by About a Year(2 posts)→
More from AGI Musings
- Scholar Discusses Performative Risks of AI Fact-Checking — chris_bail · 2026-08-27
- Elon vs. Anthropic: How Builders' Worldviews Will Shape the Future of AI — XFreeze · 2026-08-27
- Cheap Chinese AIs threatening frontier labs is a 'lump of labor fallacy,' argues Theo Jaffee — robleclerc · 2026-08-27
- A 'parsimonious' alignment fix: Urbit/Bitcoin-style hierarchical identity and auditable capital flows — curious_vii · 2026-08-27
- David Patterson on Universal High Income: Distributing Machine Output — davidpattersonx · 2026-08-27
- Universal High Income, Not Just UBI, for the Age of AI Superabundance — davidpattersonx · 2026-08-27