Linus Torvalds
Pet peeve of the day: all the people talking about how ChatGPT is not “conscious” and how it does not “understand” what it is saying, but just putting likely-sounding words together into likely-sounding sentences.
Extra bonus points for using an example of a math problem as a way to show how these AI chat-bots talk about things they don’t really understand.
The irony. The lack of self-awareness. It burns.
Linus Torvalds
Please also tell me all about “the quantum mind” or how the machine does not have a “soul”.
Linus Torvalds
@Inginsub that’s not what I claim. I claim that the people talking about “understanding” do not themselves have the very understanding that they are taking about.
They are literally just doing the same thing they talk about ChatGPT doing: putting likely words into sentences that sound good.
Smarter people than me and you have gone down crazy rabbit holes in this area for centuries.
:ihavenomouth:
Linus Torvalds
@missingno you probably say that jokingly, but I think that’s actually the real underlying truth.
It’s obviously true of ChatGPT.
But I really do think the much more interesting truth is that it’s probably true of us too.
Missingno.
Wolf480pl
@torvalds @Inginsub
I think there's a significant difference between how ChatGPT is trained and how humans are trained.
When humans learn, they're not just rewarded for saying plausibly-sounding things, but for saying useful things, and for getting things right. There's often a teacher, or a physical process one's trying to master, that has rather strict criteria of what is correct and what's wrong.
Meanwhile ChatGPT seems to be optimized for making non-experts believe it.
Disinformation Purveyor :verified_think:
Linus Torvalds
@rook it’s easy and probably largely pointless to criticize the consciousness of AI models. You’re inevitably just making stuff up, since you control the very definition of what “consciousness” is to you.
That’s kind of my point. I suspect that ChatGPT could write a decent paper on this very thing.
The much more interesting thing is to see what those models tell us about ourselves, using hard data from AI models. But I suspect there’s a lot of people who are very invested in discussions about qualia and experiences who really don’t want to go there.
Urist McTowerbird
@torvalds If you're serious and not just venting, I suspect Karl Friston has a lot to say about that. At least in the sense that there is a qualitative difference that can be quantified. But using quotes around the C-word sounds like reserving license to move the goalposts.
That said, *I* do mean something specific and quantifiable when I criticize the consciousness of ML models, and it's based in Friston's work.
０７６萌ＳＮＳ is a social network, courtesy of ０７６. It runs on GNU social, version 2.0.2-beta0, available under the GNU Affero General Public License.
All ０７６萌ＳＮＳ content and data are available under the Creative Commons Attribution 3.0 license.