So this paper found you can cut the API token costs of using Chain of Thought prompting by over 20% with no decrease in accuracy for GPT-4 (though a decrease in math accuracy in GPT-3.5) by just adding the words "be concise." That's all. LLMs are weird. arxiv.org/pdf/2401.05618 pic.twitter.com/ijPa4knVWE

What should an AI's character be? Read our post on how we approached shaping Claude’s character: www.anthropic.com/research/claude-character

Science in most domains (except mathematics) is not intelligence-bound, but experimentation-bound. You see this in AI in particular. Our ability to come up with clever ideas far outpaces our ability to test those ideas and iterate.

The ultimate AI assistant: GPT-4o, Claude, Gemini, and Llama 3 in one place. Plus, you can create your custom assistant. We built it. It's faster, more comprehensive, and cheaper than your ChatGPT subscription. You can try it for free: