Well, two responses I have seen to the claim that LLMs are not reasoning are:
So I think this research is useful as a response to these, although I think “fuck off, promptfondler” is pretty good too.
I assume it’s the same dorks who say “ChatGPT is useful to summarize emails.”
Everybody knows that all languages derive from ULTRAFRENCH.
So Geoffrey Hinton is a total dork.
Hopefully, [this Nobel Prize] will make me more credible when I say these things really do understand what they’re saying. [There] is a whole school of linguistics that comes from Chomsky that thinks it’s nonsense to say these things understand language. That school is wrong. Neural nets are much better at processing language than anything produced by the Chomsky school of linguistics.
https://www.bbc.com/news/articles/c62r02z75jyo
It’s going to be like the Industrial Revolution - but instead of our physical capabilities, it’s going to exceed our intellectual capabilities … but I worry that the overall consequences of this might be systems that are more intelligent than us that might eventually take control
😩
“I only had this problem because I was very reckless,” he continued, "partially because I think it’s interesting to explore the potential downsides of this type of automation. If I had given better instructions to my agent, e.g. telling it ‘when you’ve finished the task you were assigned, stop taking actions,’ I wouldn’t have had this problem.
just instruct it “be sentient” and you’re good, why don’t these tech CEOs undersand the full potential of this limitless technology?
This particular bit of news has me so down. :(
https://xcancel.com/karpathy/status/1841848120897912967#m
jfc
Edit: replaced screenshot of tweet + description with link to tweet
There’s vanishly little that LLMs are actually being used for that can’t be done far cheaper (computatiomally and cost-wise) with existing tools.
Reminds me of the most recent Adam Conover podcast. He had as guests two computer scientists who were purportedly critical of AI, and one of them still shat out something to the effect of:
It does have a use case where something takes longer to produce than it does to verify. For example, a website …
Every few years there is some new CS fad that people try to trick me into doing research in — “algorithms” (my actual area), then quantum, then blockchain, then AI.
Wish this bubble would just fucking pop already.
go gatekeeper somewhere else
Me, showing up to a chemistry discussion group I wasn’t invited to:
Alchemy has valid use cases. If you want to be pedantic about what alchemy means, go gatekeep somewhere else.
You’re absolutely right, unless you actually read it.
The DOJ suspects Nvidia has been:
- Pressuring customers to buy only Nvidia chips if they want continued supply;
- Buying up competitors, such as Israeli startup Run:ai, to suppress their technologies;
- Locking customers in with its software, such as the CUDA libraries.
Could it be because a statistical relation isn’t the same as a semantic one? No, I must be prompting it wrong. I’ll just add “engineer” to my title and then everyone will take me seriously.
If I were unable to write a novel in a month1 but really wanted to and some smug little shit came up to me and offered to ghost write it for me, I would not be happy. How is SALAMI generated text any different?
1: I definitely can’t.
Thanks for sharing this. <3
Oh god is this the first time we have to sneer at a 404 article? Let’s hope it will be the last.
My intention was more to sneer at the research:
https://www.404media.co/this-is-doom-running-on-a-diffusion-model/
We can boil the oceans to run a worse version of a game that can run at 60fps on a potato, but the really cool part is that we need the better version of the game to exist in the first place and also the new version only runs at 20fps.
“we are likely… fact” is just a special kind of stupid
yikers