logoalt Hacker News

GalahiSimtam • today at 9:16 AM • 9 replies • view on HN

I am old enough to remember what AGI meant before the rise of LLMs, and it used to be more or less what this article calls "ASI".

You might find more people believing that AGI is already solved. For example, the author of "The AGI chronicles" drops this in the epilogue. For him, AGI is achieved, because it can help him with a wide range of subjects, with the ability of a random stranger. It's clear that he means chatbots, internet surveys summaries, and coding bots - he does not even mention self-driving cars there.

This is a shift from the earlier chapters, where AGI term is employed somewhat more vaguely, and even quipping once that AI researchers think of AGI as a slightly smarter version of themselves.

I too, have a lower bar for AGI. It would be enough for me if it could design games I work on. Merely speeding up implementation tasks is not that interesting. In the immortal words, a delayed game is eventually good, but a rushed game is forever bad.


Replies

lordnacho • today at 1:45 PM

> Merely speeding up implementation tasks is not that interesting.

This allows you design things faster, though. More effort can be spent on iteration when you aren't fiddling with libraries and syntax. I find that I am able to discover the open questions a lot faster because you reach the decision point sooner when the machine is simply doing the trivial things.

As for the design aspects, it is pretty good at brainstorming. Whatever your choice is, I find that it throws out an exhaustive vanilla list for whatever it is. It can also critique whatever you come up with.

So for me, I also feel like AGI is already here. There's probably a few things it can't do as well as a real random person, but there's a heck of a lot of things that it does better.

➕ show 1 reply
jaapz • today at 9:52 AM

> but a rushed game is forever bad.

Is it though? CDPR famously released Cyberpunk rushed and in a miserable state. But the game is amazing now, and this is a widely held opinion too.

➕ show 7 replies
spacedcowboy • today at 10:39 AM

> I am old enough to remember what AGI meant before the rise of LLMs, and it used to be more or less what this article calls "ASI"

Me too, my first neural network research was for my PhD, 35 years or so ago, on a Meiko computing surface.

My bar for AGI over what we have (generative AI or genAI) is still fairly high, fortunately I do not believe that the current genAI path is set to lead us to AGI, there are very fundamental differences.

Right now for genAI, the model is static, the weights do not change, all “memory” is in the context, and even with the context sometimes exhibiting “J-space” (a shared "workspace" for deliberate, multi-step reasoning spontaneously created in the context memory), its ephemeral; it doesn't inform the weights in the base model. Even when we try to do “weight-space programming” to actually modify the model itself, the supervisor-model needed to do that is far and away larger than the one being operated on, and once a change has been made the supervisor-model is now out of sync with its original representation and it becomes less useful the more it diverges. We have not solved "thinking" today, we “just” have a high-probability path between states in a static weight-space which we follow and feed back.

What thinking requires is the ability to generate hypotheses (which requires internal state modification), to hold them and test them against other facts and hypotheses, and to update the internal state on test results, negative or positive. LLM’s don’t do that, they’re actually more like a database lookup, where the input tokens act as keys into a high-dimensional space where information is stored. You could even consider “the model” to be just an entropic compression of lots of data, and the inference step to be the lookup. It’s the scope, and the fuzziness of the lookup that makes them seem to work so well. Lots of results come back, and the best-path through those results wins. It’s not “thinking” as we understand it.

This is why humans are still a crucial component: we set the boundary conditions of these lookups, we set the starting point and the goals, and we adjust the success criteria to guide the chat - note that these are all the comprehension parts of the task. It’s also why models will hallucinate: the fundamental design does not lend itself to a null hypothesis. Even in the newest "reasoning" models, there is still no feedback of state back into the model itself. What they are effectively doing is running a Monte Carlo pass through the problem space—generating hidden branches and picking a likely path—before reporting back. Because the underlying weights never update from that process, it’s still just a sophisticated search, not genuine comprehension - there is still no “that makes no sense, reconfigure” step.

Stuart Kauffman (“The Origins Of Order”) would say its weights are stuck in the frozen realm of order, and that it takes a genuine phase transition to get to where the real action is - at the boundary of chaos and order where humans reside. Tweaking and fiddling with the system while it remains locked in the stasis of static weights is never going to effect that state change.

That’s not to say genAI is not useful, it’s freaking awesome, possibly the best invention in the history of time, but it doesn’t truly “think”, it maps.

➕ show 5 replies
paxy • today at 11:54 AM

My personal definition is:

AGI = as good as most humans at most tasks

ASI = significantly better than all humans at all tasks

➕ show 2 replies
RunSet • today at 1:46 PM

The lowering of the bar for "AGI" reminds me of the time I argued with a friend to nearly the limit of my patience about whether it was possible to build a perpetual motion machine, and then at last they revealed that- to them!- a perpetual motion machine was a device that could operate for the rest of their life without recharging or exhausting its power source.

LogicFailsMe • today at 2:35 PM

Agreed, AGI went from a 24/7 Stephen Hawking/Albert Einstein duet to a flaky high-latency GenZ Homer Simpson so the AI boosters could declare victory.

My bar for AGI is lower as well: I'll believe when I can give a model a complex out of sample task and instead of blindly implementing the task as I indequately described, it asks me questions to resolve ambiguity sufficiently that it builds exactly what I wanted without imposing strange semantic constraints on itself or missing braindead obvious to a junior engineer defects in the design or implementation.

➕ show 1 reply
zobzu • today at 11:17 AM

it's just marketing. the asi will be solved by some other acronym won't. and neither will be what they say they sell: actual sentience

Silamoth • today at 3:26 PM

Your first paragraphs echo my thoughts very closely. I’m not that old. But just a few years ago, everyone technical knew “AI” was really just a buzzword. Sure, you might tell investors or grant funders that you’re using “AI” to solve a problem because that sounded fancy and impressive. But whatever you were doing was fancy math and statistical modeling. Neural networks were loosely inspired by brains, but they’re really just fancy math.

The term AGI arose to distinguish these buzzword/lay uses of “AI” from ‘actual’ intelligence. Even that notion was always ill-defined and really a philosophical problem as much as a technical one. But the vision was always mimicking humans or smarter. Not just in fact recall. Not just in reading faster than humans. But logical and emotional intelligence on the level of human geniuses, accelerated by computation. Think Jarvis in Iron Man, Data in Star Trek, or C-3PO in Star Wars. You should be able to hold a natural conversation with an actual AI. You should be able to trust it to accomplish essentially any task (maybe with some instruction, like a human). It should be able to learn.

Over the last couple years, though, these tech bros seem to have redefined AI and AGI to mean LLMs - obviously because it benefits them. They are a far cry from anything described before as genuinely real AI - or AGI. I understand how nontechnical laypeople can be confused. But technical people should know better. And yet so many have fallen for the propaganda, even here on HN.

Note: This is largely orthogonal to whether or not LLMs are a useful tool

juped • today at 9:59 AM

Artificial "General" Intelligence has to be general; this term is from the era when the frontier of AI was "good, specifically, at playing Go" or something. Any robot apocalypse connotations are limited to the robot apocalypse fandom.

➕ show 2 replies