We're not even close to RSI yet. Without human intervention, current models will still spin in circles reverting to the mean. This can be sufficient for most work though, as the mean for AI models is seemingly well above that of the average person. But genius still eludes it. Sure mathematics can be proved, and data can be mined for novel proteins, etc. But where are the AI symphonies? Where's the AI Great American Novel? Mediocrity machines is all they are, yet most of us are (by definition) mediocre, so it doesn't bode well.
"Artificial general intelligence (AGI) is a hypothetical type of artificial intelligence that matches or surpasses human capabilities across virtually all cognitive tasks.[1] Unlike artificial narrow intelligence (ANI), whose competence is confined to well‑defined tasks, an AGI system can generalise knowledge, transfer skills between domains, and solve novel problems without task‑specific reprogramming."
I would argue that its not that relevant anymore what specific AGI definition you follow. The impact is already significant and real, progress hasn't stoped and the current chatbots can do arbitrary things without task-specific reprogramming.
Depending on the context size and your problem, i can also teach an chatbot new things adhoc without reprogramming.
> Artificial general intelligence (AGI) is a hypothetical type of artificial intelligence that matches or surpasses human capabilities across virtually all cognitive tasks.
Programmers think that because a slick piece of software is really good at writing code, it must obviously be really good at everything else. Nothing is more complex than the big brain shit they've been doing their whole lives.
It is inordinately funny to me that the examples you listed all involve the computer as a medium – the kind of things a programmer would think is marvelous and proof of AGI, but falls squarely within the wheelhouse of computer whiz shit.
“solve novel problems without task‑specific reprogramming."
Computer whizzes keep building benchmarks, training sets and harnesses to get it to do computer whiz shit. No amount of that will lead to general intelligence. It’s like trying to construct complex numbers by adding large lists of integers.
Right, but, again, humans micro-managing the process and aiding it with harness tools is not AGI by the very definition you quoted.
Get one to write programs in a new programming language with novel syntax and semantics that are not represented in its training set, and that might more compelling. Because that demonstrates real knowledge transfer and skills acquisition without task-specific reprogramming.
Heck, even an existing but obscure and unusual one with very little representation in public data sources would be an interesting demonstration. For example there’s CPL, an application-specific scripting language which does have a C-like syntax but lacks structs and its only dynamically sizeable data structure is a B-tree with a pretty rough-around-the-edges interface. Take an off the shelf model and demonstrate it doesn’t already know how to write it but can automatically improve at the task without undue human assistance, and you’d have a plausible demonstration of solving a (semi-) “novel problem without task-specific reprogramming.”
It’s that sort of thing that gets you the G im AGI. Not people continuing to play task whack-a-mole. The G is supposed to explicitly exclude those kinds of tricks.
Fable and Astra level models have become relatively good at these sorts of things, have not tried it with CPL but Lisp and Fortran which I doubt there is much of a significant training body on in the corpuses, but they perform extremely well. I think scaling has reached the point that these sort of things are just part of the emergent behaviors of AI
>"It can use blender and build you a 3d model of a 2d picture."
But can it do it as well as an average professional?
>"It can play a digital piano, it can make the piano first and then play it."
Can it play something you'd put on at a party and go "this is my jam!"?
The answer is no to all of the above. It can do those things, but only to a degree of competence that a very novice human would. Maybe some day in the future it won't. But they are extremely mediocre today precisely because RSI is not yet possible. They have no taste, and can only approach the median of their weights.
ramesh31 · · focus · HN ↗
Glemmlko · · focus · HN ↗
"Artificial general intelligence (AGI) is a hypothetical type of artificial intelligence that matches or surpasses human capabilities across virtually all cognitive tasks.[1] Unlike artificial narrow intelligence (ANI), whose competence is confined to well‑defined tasks, an AGI system can generalise knowledge, transfer skills between domains, and solve novel problems without task‑specific reprogramming."
I would argue that its not that relevant anymore what specific AGI definition you follow. The impact is already significant and real, progress hasn't stoped and the current chatbots can do arbitrary things without task-specific reprogramming.
Depending on the context size and your problem, i can also teach an chatbot new things adhoc without reprogramming.
nozzlegear · · focus · HN ↗
Programmers think that because a slick piece of software is really good at writing code, it must obviously be really good at everything else. Nothing is more complex than the big brain shit they've been doing their whole lives.
Glemmlko · · focus · HN ↗
It can play a digital piano, it can make the piano first and then play it.
It can build 3d simulation based on research paper.
It already surpasses plenty of 'big brains' on our whole planet.
nozzlegear · · focus · HN ↗
Glemmlko · · focus · HN ↗
We now have a system which can use tools and you still say this is just 'computer whiz shit'?
A computer is a tool, a very generic one which we adopted to control our most critical systems.
Enlighten me, when do you become impressed?
bunderbunder · · focus · HN ↗
Computer whizzes keep building benchmarks, training sets and harnesses to get it to do computer whiz shit. No amount of that will lead to general intelligence. It’s like trying to construct complex numbers by adding large lists of integers.
Glemmlko · · focus · HN ↗
They then either tell someone to execute it in real world or will upload this to a robot.
Nvidia has already the Omniverse for this and others are working on World Models.
bunderbunder · · focus · HN ↗
Get one to write programs in a new programming language with novel syntax and semantics that are not represented in its training set, and that might more compelling. Because that demonstrates real knowledge transfer and skills acquisition without task-specific reprogramming.
Heck, even an existing but obscure and unusual one with very little representation in public data sources would be an interesting demonstration. For example there’s CPL, an application-specific scripting language which does have a C-like syntax but lacks structs and its only dynamically sizeable data structure is a B-tree with a pretty rough-around-the-edges interface. Take an off the shelf model and demonstrate it doesn’t already know how to write it but can automatically improve at the task without undue human assistance, and you’d have a plausible demonstration of solving a (semi-) “novel problem without task-specific reprogramming.”
It’s that sort of thing that gets you the G im AGI. Not people continuing to play task whack-a-mole. The G is supposed to explicitly exclude those kinds of tricks.
Avid_F · · focus · HN ↗
ramesh31 · · focus · HN ↗
>"It can use blender and build you a 3d model of a 2d picture."
But can it do it as well as an average professional?
>"It can play a digital piano, it can make the piano first and then play it."
Can it play something you'd put on at a party and go "this is my jam!"?
The answer is no to all of the above. It can do those things, but only to a degree of competence that a very novice human would. Maybe some day in the future it won't. But they are extremely mediocre today precisely because RSI is not yet possible. They have no taste, and can only approach the median of their weights.