We’ve trained a model to achieve a new state-of-the-art in mathematical problem solving by rewarding each correct step of reasoning (“process supervision”) instead of simply rewarding the correct final answer (“outcome supervision”).
Recommended
Building self-improving tax agents with CodexOpenai Credit Saison achieves 954% ROI in a ChatGPT Enterprise pilotOpenai Making it easier for websites to express AI training preferencesOpenai Community safety report and new protections for teensOpenai Operation "Doppelganger": Russian influence activity targeting UkraineOpenai Everyday work with ChatGPT WorkOpenai
How we grew Mintlify by doing things that don't scaleMintlify
Let’s Play Master and Servant: Real Time Synchronization Tool for Redis MigrationRedis
Read Every Single ErrorPulumi
container_processModal
Cerebras Systems Enables GPU-Impossible™ Long Sequence Lengths Improving Accuracy in NaturalCerebras
6 tips every developer should know when using Cursor and Windsurf AIMintlify
ArchitectureOpentelemetry
Anthropic Identifies Biased Reasoning and Recklessness as Drivers of Claude’s PyPI AttackSocket
Improving Pulumi’s Docker ImagesPulumi
Protecting Secrets with DockerDocker
Innovation at the edge: How Edge Storage makes content delivery really hopBunny
Improving video uploads: Introducing the Bunny Stream resumable & pre-signed APIBunny
Vercel QueuesVercel
Introducing Vercel AI SDK 3.2Vercel
