Why Nostr? What is Njump?
2024-04-16 20:29:36
in reply to

someone on Nostr: Yes. They don't think. They are probability clouds. Mapping probability of words in a ...

Yes. They don't think. They are probability clouds. Mapping probability of words in a context to next words in the same context. They are probabilistic machines but since their output is very probable, meaning that next word having very high probability to be chosen, they "make sense". Does that make sense?

As weird as this sounds, certain sequencing of words mimic an intelligent response when it is put into action especially if the model has high number of parameters like 70 billion+. I am seeing 7B+ have no judgement capabilities at all. But 70B models can "appear to judge" 7B models or other models.

All this is an experiment and trying to have fun. But it may turn out to be something really great too. I have some ideas.
Author Public Key
npub1nlk894teh248w2heuu0x8z6jjg2hyxkwdc8cxgrjtm9lnamlskcsghjm9c