AI models' human-equivalent time horizons scale sub-linearly with tokens, per UK AISI
tobyordoxford · x · 2026-10-09
Toby Ord highlights a chart from UK AISI showing that recent AI models' human-equivalent time horizons scale sub-linearly with tokens — more training yields diminishing gains in how long tasks models can handle.
He adds that finer-grained human measurements suggest the human curve bends upwards rather than downwards in that range, an important reference point for model-vs-human scaling comparisons.
Related event: UK AISI Data Shows Model Time Horizons Scale Sublinearly with Tokens(4 posts)→
More from Research
- RNA fingerprinting method published in Cell infers drivers of cellular responses from Perturb-seq — anshulkundaje · 2026-10-10
- Team builds 8,000 sqft wet lab in 30 days; AI-generated compounds kill $13B pest Fall Armyworm — aadityabuilds · 2026-10-10
- Snorkel Scales Open Benchmarks Grants to $30M as Marin Lead Explains How Evals Guide Training — dlwh · 2026-10-10
- Computational biologists urge reconstructing key biological axes in latent space — anshulkundaje · 2026-10-10
- Success-Guided Sampling: sim-to-real RL nails dexterous assembly with zero demos — kevin_zakka · 2026-10-10
- AI-for-science debate: phenotype readouts matter more than gene-by-gene reconstruction — fabian_theis · 2026-10-10