longwriter

Current long-context large language models (LLMs) can process inputs as much as 100,000 tokens, yet they struggle to generate outputs exceeding even a modest length of two,000 words. Controlled experiments reveal that the model’s...