2 posts tagged model-performance.
OpenAI signs Pentagon deals while Anthropic fights back, plus new research shows even frontier models decay in long conversations.
Every speedup technique is just teaching models to guess better, and we're running out of good guesses.