OpenAI has reportedly (per @synthwavedd ) finished training “Bel,” a massive successor to “Doug” with over 10 trillion total parameters, probably the base for GPT-6.
But more interesting: Anthropic reportedly lacks the compute to answer OpenAI’s upcoming Astra launch this year.
Project Stargate is apparently paying off, but not in the way expected. Not so much because they can now handle more inference, but because the compute capabilities will allow them to train significantly better models in the long term.