all the “how Jev works” videos and articles are actually “how to use it”
Here’s how it works: Jev is an LLM that predicts ONE TOKEN.
Thats it. Thats all it is.
Correction: “one out of the 2-3 you suggest”
all the “how Jev works” videos and articles are actually “how to use it”
Here’s how it works: Jev is an LLM that predicts ONE TOKEN.
Thats it. Thats all it is.
Correction: “one out of the 2-3 you suggest”
Jev for dummies
Classic LLMs answer one question "Whats the next token". Then add this token back to your text and repeat.
@typesafeai works backwards. It says "*GIVEN* the next token is A or B - tell me the probabilities"
U're welcome
P.S. Also look up "zeroshot" on h.face
And no, I odn't think this is diffusion based. No idea how it works, maybe calculates the quesitons's embeddings then calculates distance to both answers? Well, that just makes it a more complex RAG then... Anyway, this is interesting as hell
Keep coming back to that Miro+Bending Spoons deal... Only 2.3X ARR...
I still don't think it's "the death of SaaS". Just a reminder that late stage VC-backed startups are expected to have insane growth. Otherwise you get a brutal multiplier.