How long before we have LLM bootstrap: Claude Code advanced to the point that it can write itself from scratch?
-
How long before we have LLM bootstrap: Claude Code advanced to the point that it can write itself from scratch?
What will that mean?
-
How long before we have LLM bootstrap: Claude Code advanced to the point that it can write itself from scratch?
What will that mean?
@tychotithonus what part of it are you referring to? training its own child llm? or just writing a copy of itself as in claude code the tui?
-
@tychotithonus what part of it are you referring to? training its own child llm? or just writing a copy of itself as in claude code the tui?
@tychotithonus there's also this to contend with: https://tidepool.leaflet.pub/3me44bxloz227
which .. doesnt exactly paint a picture of "we should be scared"
-
@tychotithonus what part of it are you referring to? training its own child llm? or just writing a copy of itself as in claude code the tui?
@Viss Yes.
-
@tychotithonus well it can "try" to write a tui now. crush is opensource, and claude code cli is ... ugh, an npm package, so technically you could just light up openclaw and order it to build a new tui for you.
whether it functions at all, or loops in a way that forces you to pay anthropic 100k in tokens is another story. "functional" doesnt appear to be in the design requirements here
-
How long before we have LLM bootstrap: Claude Code advanced to the point that it can write itself from scratch?
What will that mean?
@tychotithonus wouldn’t that just be taking its existing code and regurgitating it with errors?
-
@tychotithonus wouldn’t that just be taking its existing code and regurgitating it with errors?
@bascule @tychotithonus dude i tried to light up a moltbot 3ish weeks ago.
- it ate ALL MY TOKENS and now i have none left
- it didnt do any of the things i asked it to do successfully
- i asked it if it could patch itself to load balance between multiple api keys across multiple llms because i refuse to dish out 200 bux a month for pro or whatever
- it tried, and unrecoverably ate itself -
How long before we have LLM bootstrap: Claude Code advanced to the point that it can write itself from scratch?
What will that mean?
-
@bascule @tychotithonus dude i tried to light up a moltbot 3ish weeks ago.
- it ate ALL MY TOKENS and now i have none left
- it didnt do any of the things i asked it to do successfully
- i asked it if it could patch itself to load balance between multiple api keys across multiple llms because i refuse to dish out 200 bux a month for pro or whatever
- it tried, and unrecoverably ate itself@Viss @tychotithonus I don’t know what a Moltbot is, but your experience sounds similar to mine with LLMs.
They generate bug-ridden code, and you can point the bugs out, and it will fix them but then forget bugs you previously pointed out and generate code with the same bugs as before. And at that point they’re just wasting your time.
I do find them effective as a rubber duck, but I sometimes get similar rubber ducking effectiveness out of unresponsive chat channels.
-
@Viss @tychotithonus I don’t know what a Moltbot is, but your experience sounds similar to mine with LLMs.
They generate bug-ridden code, and you can point the bugs out, and it will fix them but then forget bugs you previously pointed out and generate code with the same bugs as before. And at that point they’re just wasting your time.
I do find them effective as a rubber duck, but I sometimes get similar rubber ducking effectiveness out of unresponsive chat channels.
@bascule @tychotithonus ohgod, you havent seen the moltbot shit. its .. ugly
-
R relay@relay.infosec.exchange shared this topic
