Skip to content

Reasoning

Progress in model reasoning: chain of thought, reasoning models, math and logic benchmarks and the debates around them.

Latest picks

701–705 of 705

Sep 12, 2024Thursday

OpenAI News

Learning to reason with LLMs

OpenAI released o1-preview and reported 74% single-sample accuracy on AIME 2024, versus 12% for GPT-4o. The post says o1 reached the 89th percentile on Codeforces and exceeded human PhD experts on GPQA Diamond; it attributes this to large-scale RL and gains from both train-time and test-time compute. The key signal is scaling reasoning with compute, not just pretraining a larger base model.

Why it matters: OpenAI's o1-preview scores 74% on AIME 2024 math problems against GPT-4o's 12%, rising to 83% with repeated sampling.

OpenAI News

OpenAI o1-mini

OpenAI released o1-mini on Sept. 12, 2024 for Tier 5 API users at 80% lower cost than o1-preview. The post reports 70.0% on AIME and 1650 Codeforces Elo, close to o1 at 74.4% and 1673, with about 3-5x faster answers than o1-preview in one word-reasoning example. The key tradeoff is explicit: it targets STEM reasoning, while non-STEM factual knowledge is only comparable to small models like GPT-4o mini.

Why it matters: OpenAI's o1-mini costs a fifth of o1-preview per API call, opening first to paying ChatGPT and Tier 5 API users, so the price-capability tradeoff is the key detail.

Jul 17, 2024Wednesday

OpenAI News

Prover-Verifier Games improve legibility of language model outputs

OpenAI trained GPT-4-family prover-verifier games so stronger models write solutions weaker models can verify; under time-limited human review, correctness-only optimization led to nearly 2x more evaluation errors. The post says the large and small models differ by about 3 orders of magnitude in pretraining compute, and checkability training recovers about half the performance gain of correctness-only optimization; the full experimental numbers are not fully disclosed in the provided text.

Why it matters: Training a large prover against a small verifier cut human evaluation error from double to half, offering a training pattern that transfers to other scalable oversight problems.

Jun 28, 2024Friday

DeepSeek · API updates

deepseek-chat upgraded to DeepSeek-V2-0628, with better reasoning and role-play

The deepseek-chat model has been upgraded to DeepSeek-V2-0628. Reasoning improved, and role-play got a clear boost. HumanEval Pass@1 rose from 79.88% to 84.76%, MATH ACC@1 from 55.02% to 71.02%, and BBH from 78.56% to 83.40%.

Why it matters: The math benchmark gain is larger than the code and reasoning gains, and the win rate against GPT-4-0314 on Arena-Hard also improved.

Jun 14, 2024Friday

DeepSeek · API updates

deepseek-coder upgraded to DeepSeek-Coder-V2-0614

The deepseek-coder model has been upgraded to DeepSeek-Coder-V2-0614, with a clear gain in coding ability. The company says its code generation, code understanding, code repair and code completion reach the level of GPT-4-Turbo-0409, with strong math and reasoning, while general ability matches DeepSeek-V2-0517.

Why it matters: DeepSeek upgrades its code model and benchmarks code generation, understanding, repair and completion against GPT-4-Turbo-0409.