entity watching

test-time compute

Watching stub, created early per Cali ruling 2026-07-07 — the ruling-7 re-source topic (Epoch AI / Stanford HAI primaries) is expected to grow this cluster. Volatile term; watch for it displacing "inference scaling" or being displaced.

References

claim-test-time-compute-can-substitute-parameters · claim-extended-thinking-as-serial-inference-compute · claim-inference-dominant-ai-compute-2026 · moc-inference-economics · question-bridge-training-inference-compute
2026-10-21: the pre-industrial lineage — anytime algorithms (Dean & Boddy 1988), Horvitz's 1987 real-time rationality, and bounded optimality — claim-dean-boddy-1988-anytime-algorithms-formalize-compute-for-quality-tradeoff · claim-anytime-algorithms-continuously-connected-not-stent-premature

· raw markdown

Linked from

Does AI/cognitive-science history contain a genuinely Stent-premature articulation of inference-time ('think longer at answer-time') compute?

capture
Journal — 2026-10-21

journal
Dean & Boddy's 1988 paper named 'anytime algorithms' and formalized trading extra computation at answer-time for a better answer — but not first: its own footnote credits Horvitz's more general 1987 framework

claim
Anthropic describes extended thinking as 'serial test-time compute': sequential reasoning steps before final output

claim
s1 reproduced o1-preview-level math reasoning by fine-tuning Qwen2.5-32B on 1,000 distilled traces in 26 minutes

claim
Scaling test-time compute optimally can enable a smaller model to outperform one 14x larger

claim
anytime algorithm

entity
Chain-of-thought prompting

entity
Eric Horvitz

entity
Niklas Muennighoff

entity
o1

entity
Thomas Dean

entity