5 ms·

I wonder how much of Anthropic's revenue comes from tokens saying "you're absolutely right!"

• subscribed 1y ago
  "You're concise" in the "personality" setting saves so much time. Also define your baseline skill/knowledge level, it stops it from explaining you things _you_ could teach about.
• alentred 1y ago
  Oh wow, I never thought of that. In fact, this surfaces another consideration: pay-per-use LLM APIs are basically incentivized to be verbose, which may be well in conflict with the user's intentions. I wonder how this story will develop. In an optimistic sci-fi line of thinking, I would imagine APIs using old-school telegraph abbreviations and inventing their own shortened domain languages. In practice I rarely see ChatGPT use an abbreviation, though.
  • SJMG 1y ago
    > pay-per-use LLM APIs are basically incentivized to be verbose There's competing incentives. Being verbose, let's them charge for more tokens, but it's also not prized by text-consumers in the most common contexts. As there's competition for marketshare, I think we see this later aspect dominate. Claude web even ships with a "concise" mode. Could be an issue long term though, we'll have to wait and see! > In an optimistic sci-fi line of thinking, I would imagine APIs using old-school telegraph abbreviations and inventing their own shortened domain languages. In the AI world this efficient language is called "neuralese". It's a fun rabbit hole to go down.
• SJMG 1y ago
  The smart inversion of saying "thank you" costing OpenAI millions https://www.vice.com/en/article/telling-chatgpt-please-and-thank-you-costs-openai-millions-ceo-claims/ https://www.vice.com/en/article/telling-chatgpt-please-and-t...