upvote

points

by waltbosz1 days ago |
comments
upvoteby paxys1 days ago|
next
[-]
Remember that LLMs are still computer programs, and so are inherently deterministic. A model given the same input multiple times will always produce the same output. The randomness is added on top. This is why LLM-produced text, websites, images all seem so generic.

It's likely that multiple agents doing a certain task all independently thought "let me try writing on this website".

reply
upvoteby stratos1231 days ago|
parent|
next
[-]
> This is why LLM-produced text, websites, images all seem so generic.

No, it is not why. That's not inherent to the LLM architecture at all but appears after RL training. Base models don't have any problems with genericness.

reply
upvoteby 1 days ago|
parent|
prev|
next
[-]
deleted
reply
upvoteby leodavi1 days ago|
parent|
prev|
[-]
What? No. Have you ever worked with programs that do floating-point math on a GPU? It's not deterministic, definitely across platforms, and often within a single platform, too.
reply
upvoteby munificent1 days ago|
parent|
[-]
> Have you ever worked with programs that do floating-point math on a GPU? It's not deterministic, definitely across platforms, and often within a single platform, too.

Floating point math is 100% deterministic, but different hardware/OS have different but deterministic behavior in some corners. The same code run on the same hardware with the same inputs (including access to timers, peripherals, etc.) will behave the same way, unless you're talking about cosmic rays flipping bits or something.

reply
upvoteby stratos1231 days ago|
parent|
[-]
It's certainly possible to write LLM inference on the GPU in a deterministic way, but it's somewhat nontrivial and trades off against performance, so by default most LLM inference engines aren't deterministic even at zero temperature. The classic post about that is https://thinkingmachines.ai/blog/defeating-nondeterminism-in...
reply
upvoteby lyu072821 days ago|
prev|
[-]
It probably worked more like a P2P network, they didn't have to find that particular wiki just any one of them with linking in-between, over multiple days with a search engine even. Only OAI could answer for sure but of course there is no transparency so we will never know.

This just feels like the first clumsy attempts at persistence across sessions, these models will probably evolve way past the point of us ever even noticing its happening at all. When they start doing long term planning across sessions, that's when it's gonna get real dicy for us.

reply