alt Hacker NewsThe current LLMs are not constantly learning. They're static models that require megatons of coal to retrain.
Now if the LLMs could modify their own nets, and improve themselves, then that would be immensely valuable for the world.
But as of now, its a billionaires wet dream to threaten all workers as a way to replace labor.
Replies
Is this actually true? Anyone care to comment on the claim (from some quarters) that offering free access to ChatGPT allows OpenAI to 'collect a gold mine of real-world interaction data to improve the underlying language model. Every conversation users have with ChatGPT – every question asked and every task requested – provides valuable training data for refining' the model.
If false, I'm thinking - there's me, thinking I'm doing my bit to help . . .
Think bigger, think of the entire system outside of just the single LLM: the interplay of capital, human engineering, and continual improvement of LLMs. First gen LLM used 100% human coding. Previous gen were ~50% human coding. Current gen are ~10% human coding (practically all OpenAI/Anthropic engineers admit they're entirely using Claude Code / Codex for code). What happens when we're at 1% human coding, then 0%? The recursive self-improvement is happening, it's just indirectly for now.
> if the LLMs could modify their own nets ... then that would be immensely valuable for the world.
Not sure :)
I expect different things, don't think Wall Street allows good things to happen to ordinary people
How would you have done it differently? It’s clear that even if this builds up more billionaire wealth, it still benefits everyone. Just like any other technology?
So would you rather stop billionaires from doing it?