OpenAI

OpenAI's new Rule-Based Rewards method improves AI safety without extensive human data collection. It uses simple rules to evaluate outputs, balancing helpfulness and safety in AI models.

OpenAI's GPT-4o mini offers high AI performance at 60% lower cost than GPT-3.5 Turbo. It excels in reasoning, math, and coding, with a 128K token context window.

OpenAI's 'Prover-Verifier Games' method improves AI text clarity. It trains advanced AI to create outputs that simpler AI can easily check, enhancing human understanding.

OpenAI and Los Alamos National Lab partner to study safe use of multimodal AI in bioscience labs, focusing on assisting humans with lab tasks.

Tech giants' retreat from OpenAI's board reflects growing regulatory concerns over AI partnerships and their impact on market competition.

CriticGPT, a GPT-4 model, identifies ChatGPT code errors. Human reviewers using it outperform others by 60%. It shows 63% preference in catching real bugs.