someone on Nostr: i make an llm that learns from the best people on the planet. some people are already ...

i make an llm that learns from the best people on the planet. some people are already downloading it according to huggingface. this is analogous to 'seeking truth'. my theory is if you combine good people into one llm entity even though some of them does not get everything right the resulting work will get most things right. domain experts combined will be great at finding truth.

i make a leaderboard that shows the misalignment levels of other llms. this is 'slowing down lies'. if any llm is far from truth it will rank low in the leaderboard. this work is necessary because only an llm can reach speeds of another llm to judge the outputs. humans can't possibly check llm outputs, because they are not fast enough. however an llm can easily check other llms.