Hacker Newsnew | past | comments | ask | show | jobs | submit | spartacusnacho's commentslogin

We already don't understand LLMs and they are having oberservable effects

If some LLM somewhere managed to produce a coherent and _correct_ explanation of how it worked inside, but no human ever saw it, would it matter?

We know how to apply LLms to problems, which is a subtly different thing to understanding how they do what they do. It's similar to fire: I can cook using fire, but I don't really understand how fire _works_. Heat+oxygen+fuel, sure, but what goes on chemically? I dunno. Doesn't stop me using it. (Pretty sure _humanity_ knows how fire works, though).


They also benefit from the commodification of software/knowledge work since they own manufacturing


This could be a backdoor to accomplishing just that, which I would welcome


"AI recombines successfully" is a phrase that makes me think of the human corpus as a genome and a particular AI response as a child of the parent prompt, with the thinking a model does in response to the prompt basically being a form of information sex.


That hasn't been true since the invention of the smartphone


Which in turn is bad for vibe coding


Javascript and Python have the most training data by far though, right?


"I got the feeling that Haskell was a quirky choice, t3h penguin of doom holding a spork."

I love how recognizable Millennials can be online


You (presumably) can't see it, but I'm shaking my cane at you.


I only say that because I also have a cane, shaped like epic atheist bacon :)


How the hell do you even begin without source code? Run a decompiler on the binary?


Machine code is code like any other code. Sure, often not easy to read or modify, but not impossible. I'd start with a disassembler, a debugger, possibly a decompiler and lots of patience. Given appropriate time and pay I might even consider it a fun job.


Would a disassembler be more reliable than a decompiler you think? That could be a whole lot of code you'd be looking at in assembly language.


I suppose it depends on the code you're trying to understand. Depending on its size, style (therefore, ability for a human or for a decompiler to understand it) and other factors, the best blend of tools can change.


Exactly what I’m thinking. What is there even to maintain other than restart it when it fails?


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: