此事件的相关记录
2-
AI 模型
OpenAI o1
The first reasoning model: compute moved from training to thinking.
o1 (September 2024) was trained with reinforcement learning to produce long private chains of thought before answering, jumping from GPT-4o's ~12% to ~74–83% on AIME math and expert-level science benchmarks. It introduced the test-time-compute scaling law (more thinking, better answers) that every lab immediately copied and that reshaped inference hardware…
开放记录 → -
概念
Test-time Compute & Reasoning
The 2024 discovery: models get smarter by thinking longer.
Reasoning models produce long internal chains of thought before answering, and accuracy scales with how much compute they spend thinking: a new scaling axis alongside model size and data. o1 introduced it; R1 open-sourced it; every frontier model now has a thinking mode. The infrastructure consequence is enormous: a single hard question can cost thousands…
开放记录 →