Users are discussing how "thinking efforts" or "reasoning tokens" are implemented in large language models like Claude and ChatGPT. The core question is whether these different effort levels correspond to distinct models or simply different parameter settings within the same model. One user noted that changing the effort level mid-conversation in Claude's CLI triggered a warning about breaking the cache, suggesting a more integrated system than simply adjusting parameters. AI
IMPACT Understanding how 'thinking effort' parameters function could inform prompt engineering and model behavior.
RANK_REASON User discussion and speculation about LLM implementation details.
Read on HN — claude cli stories →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →