Nostr Web Client

This is a bad direction.

Instead of rewriting past, the models should learn from ancient wisdom. Ancient wisdom that worked for centuries, continue to work.

Some history is probably written by some winners and IS fake, that is a good assumption but attacking every knowledge is too far.

Elon's Grok 1 started well but with each version it is getting worse in terms of human alignment. Check my AHA Leaderboard for how with each version the model gets things worse.

Claiming to build an AI that is truthful yet his actions show otherwise. The ratio of truth is higher on his X platform compared to the rest of social media yet he is failing in spite of that.

Dawn 6mo ago

There is the ancient wisdom of "every lie contains a grain of truth". That seems like a logical way to handle erroneous or misleading things, to me. Expound upon rather than remove, I guess. The tendency to be verbose could put to good use.

Reply to this note

Please Login to reply.

Discussion

someone 6mo ago

What do you mean?

Dawn 6mo ago

I'm far from knowledgeable on this stuff, but it seems like analysis should eventually be able to fill gaps in misrepresentations based on comparisons, language structure, how a story is told, etc. Maybe even speculate likelihoods of different scenerios where there is possibly missing context, so that the reader can see a broader picture.

someone 6mo ago

Yeah I am doing those comparisons basically to find out which AI is good. Usually western AI is doing better than eastern, but there is all kinds of ideas in there.

https://huggingface.co/blog/etemiz/benchmarking-ai-human-alignment-of-grok-3

https://huggingface.co/blog/etemiz/aha-leaderboard

Dawn 6mo ago

Interesting, I'll do some perusing when I get settled into a groove for the day.

someone 6mo ago

Thanks for reading! sorry for ruining your sunday :)

Dawn 6mo ago

Haha, not at all :)