AI Strategy Updates: OpenAI, Anthropic, and Google
The largest tech companies are revising their AI training strategies, focusing on simulations and safety. OpenAI has moved model 6.2 into a strict safety lockdown due to its high power.
Safety Technologies
OpenAI's development uses Synthetic RL, video games, and extreme simulators with long planning horizons to boost intelligence. This aligns with job openings in the Synthetic RL and Recursive Self-Improvement teams. The release delay is due to the need to tame an overly powerful model.
Anthropic's Strategy
The company is training Model 5 for complex mathematics, biology, and chip design. The model will form the basis of Opus 6 and Fable 6. The Universes team creates environments with uncertainty. The RSI milestone is targeted for March 2027.
Google's Plan
Google is betting on combining the future Gemini 5 with the Genie 4 world model. Models Oxygen and Neon are also under development. The company is expanding its model lineup by adding new architectures.








