AI 'capability' can't be reduced to a single number for forecasting, researchers argue
davidmanheim · x · 2026-09-16
David Bellamy argues 'capability' should describe what task types an AI model can do today, not serve as a hand-wavy quantity for forecasting future models. David Manheim extends the point: nearly nothing we measure is a single coherent quantity—stock indices, temperature, even physics measurements are inferences over partly understood interactions.
Related event: Researcher Argues AI Capability Cannot Be Compressed Into a Single Number(2 posts)→
More from AGI Musings
- Veteran virus engineer rebuts claims that AI lets amateurs build bioweapons — aiamblichus · 2026-09-16
- WhatsApp head: AI agents threaten to disrupt India's BPO sector and bank lending — rohanpaul_ai · 2026-09-16
- ApprenticeBench claims AI agents surpass humans on real jobs; Liang Wenfeng calls it 'prompt engineering bullshit' — ysu_nlp · 2026-09-16
- AI agents failing to coordinate across trust boundaries is a rediscovery of why institutions exist — RileyRalmuto · 2026-09-16
- You don't predict a tsunami by counting drownings: AI safety needs leading indicators — davidmanheim · 2026-09-16
- AI doomers may be America's most technophilic group, with nuclear as their one exception — ben_j_todd · 2026-09-16