Longer test-time compute and collapsing inference costs could make intelligence a dial
iruletheworldmo · x · 2026-07-22
Why longer test-time compute could make intelligence a dial
The post argues that model capability may increasingly scale with how much test-time compute you give it, citing Noam Brown’s view that longer productive reasoning can keep improving performance for weeks before plateauing.
It then combines two efficiency claims: OpenAI reportedly cut the cost of some workloads by more than half through software improvements, and a new Vera Rubin result reportedly shows 10x more DeepSeek-R1 tokens per megawatt than the previous Blackwell system.
Put together, the argument is that:
- More reasoning can buy more intelligence.
- The cost of that reasoning is collapsing.
- This could enable not just one long-thinking model, but thousands of agents running for hours, days, or weeks.
The post closes by saying we may not need to wait for giant 2029 superclusters to see unusual breakthroughs, pointing to an internal OpenAI model that reportedly autonomously disproved a longstanding Erdős unit-distance conjecture.
Related event: Extended Inference and Plunging Costs Could Accelerate ASI(3 posts)→
More from AGI Musings
- Microsoft's reported 60,000 voluntary exits spark essay asking who AI is really for — _jaydeepkarale · 2026-10-27
- PHP2Shell exploits confirmed in the wild, attackers may use agentic AI for post-exploitation — AravSrinivas · 2026-10-27
- LukeW: Technology Limits Are Temporary — and AI Skeptics Are Already Feeling It — LukeW · 2026-10-27
- GiveDirectly pilots an optional AI chatbot alongside cash aid in Rwanda — NathanpmYoung · 2026-10-27
- Ex-Anthropic employee essay: the org optimizes for power 'constrained by being seen as the good guys' — RichardMCNgo · 2026-10-27
- Investor: model version numbers are a mirage, only datacenter margins are real — StewartalsopIII · 2026-10-27