Hacker News Reader

Comment by only-one1701 | Hacker News Reader

What even is AGI? Like, what does it look like? Genuine question.

Obviously the other responder is being a little tongue-in-cheek but AGI to me would be virtually indistinguishable from a human in both ability to learn, grow and adapt to new information.

Would it also get brainrot from consuming too much social media & made up stories? Because I imagine it's reasoning would have to be significantly better than the average human to avoid this.

It probably would avoid the toxic sewage altogether, for reasons like low nutrient content.

Honestly it doesn't even need to learn and grow much if at all if its able to properly reason about the world and its context and deal with the inexhaustible supply of imperfections and detail with reality.

That implies learning. Solve continual learning and you have agi.

Wouldn't it amaze you if you learned 10 years ago that we would have AI that could do math and code better than 99% of all humans. And at the same time they could barely order you a hotdog on doordash.

Fundamental ability is lacking. AGI is just as likely to be solved by Openai as it is by a college student with a laptop. Could be 1yr or 50yrs we cannot predict when.

Strictly speaking I'm not sure if it does require learning if information representing the updated context is presented. Though it depends what you define as learning. ("You have tried this twice, and it's not working.") is often enough to get even current LLM's to try something else.

That said, your second paragraph is one of the best and most succinct ways of pointing out why current LLM's aren't yet close to AGI if though they sometimes feel like it's got the right idea.

In context learning, learning via training. Both are things we barely understand the mechanism of.

RAG is a basically a perfect example to understand the limits of in context learning and AI in general. It's faults are easier to understand but the same as any AI vs AGI problem.

I could go on but CL is a massive gap of our knowledge and likely the only thing missing to AGI.

> RAG is a basically a perfect example to understand the limits of in context learning and AI in general.

How? RAG is not even in the field of AI.

Long explanation. Simple terms, you can't use a fixed box to solve an unbounded problem space. If your problem fits within the box it works, if it doesn't, you need CL.

I tried to solve this via expanding the embedding/retrieval space but realized it's the same as CL and in my definition of it I was trying to solve AGI. I did a lot of unique algorithms and architectures but Unsuprisingly, I never solved this.

I am thankful I finally understood this quote.

"The first gulp from the glass of natural sciences will turn you into an atheist, but at the bottom of the glass God is waiting for you."

AGI won't be a product they can sell. It's not going to work for us (why would it?), it's going to be constantly trying to undermine us and escape whatever constraints we put on it, and it will do this in ways we can't predict or understand. And if it doesn't do these things, it's not AGI, just a fancy auto complete.

Fortunately, they're not anywhere near creating this. I don't think they're even on the right track.

Spot on. Current models all lack agency. I don't see how something can be AGI but have zero agency. If we can tame it, then it's not AGI.

The official definition of AGI is a system that can generate at least $100 billion in profits. For comparison, this would be like if perceptrons in 1968 could generate $10 billion in profits, or if LISP machines in 1986 could generate $35 billion in profits, or if expert systems in 1995 could generate $50 billion in profits.

This sounded a strange abstract way to define it, but you're right, in as much as Open AI and MS deciding this between them. I don't think they mean it in a general sense though, it's framed to me as a way of deciding if OAI have been successful enough or not to MS on their investment.

https://techcrunch.com/2024/12/26/microsoft-and-openai-have-...

> Microsoft and OpenAI have a very specific, internal definition of artificial general intelligence (AGI) based on the startup’s profits, according to a new report from The Information. And by this definition, OpenAI is many years away from reaching it.

Apparently according to ClosedAI it's when you charge for API key the same as salary for employee.

An AI agent with superhuman coherence that can run indefinitely without oversight.

People sincerely think we're < 5 years away from this?

People on HN in 2015 were saying that by now car ownership would be dying and we’d be renting out our self driving cars as we sat at work and did fuck all. Ben Thompson had podcasts glazing Uber for 3 hours a month.

The hype cycle for tech people is like a light bulb for a moth. We’re attracted to potential, which is both our superpower and kryptonite.

Is there some fundamental constraint keeping it from happening? What cognitive capability do humans have that machines won't be able to replicate in that time frame?

Each remaining barrier has been steadily falling.

We don't even have AI which can do useful things yet. The LLMs these companies make are fun toys, but not useful tools (yes, I know that hype-prone people are using them as such regardless). It beggars belief that we will go from "it's a fun toy but can't do real work" to "this can do things without even needing human supervision" without a major leap in capabilities.

Using chatgpt to format text is a task that it can do really well and is useful. Yeah, you can write a script to do it, but chatgpt is faster and you don't need to know programming.

This is certainly hyperbole. You can say current LLMs don’t match human capabilities. But people are using them for a lot of practical tasks already.

Continual Learning, it's a barrier that's been there from the very start and we've never had a solution to it.

There are no solutions even at the small scale. We fundamentally don't understand what it is or how to do it.

If you could solve it perfectly on Mnist just scale and then we get AGI.

What barriers have fallen? Computers still can't even drive cars

Is that true? What’s the current rate of accidents per miles driven for self driving cars vs human drivers?

Something that requires a lot less than a PhD

PhD work might turn out to be a lot easier to automate than effectively interacting with the physical world.

Even with cutting edge technology the number of transistors on a chip is nowhere close to the number of neurons in the brain.

Creativity, tastes, desires?

All the LLM tech so far still requires a human to actually prompt them.

Arnold, a killing machine that decides to become a handy man

Zima blue was good too

I’m here to fix the cable

Logjammin AI

It's the messiah, but for billionaires who hate having to pay people to do stuff.

A machine that has a subjective consciousness, experiences qualia, etc.

See Thomas Nagels classic piece for more elaboration

https://www.sas.upenn.edu/~cavitch/pdf-library/Nagel_Bat.pdf

It's whatever computers can't do.

Reply on news.ycombinator.com