4 ms·

What if we already have AGI. What if it is ChatGPT 5. What then? https://aimagazine.com/articles/openai-ceo-chatgpt-would-have-been-agi-five-years-ago https:

What if we already have AGI. What if it is ChatGPT 5. What then? https://aimagazine.com/articles/openai-ceo-chatgpt-would-have-been-agi-five-years-ago https://aimagazine.com/articles/openai-ceo-chatgpt-would-hav... Edit: this was serious, if I read the Wikipedia definition of AGI, ChatGPT meets the historical definition at least. Why have we moved the goal posts?

junon 1y ago
Sam is wrong here, AGI has never had a clear definition, this there's no "we have" or "we haven't". I'd say most agree we're still a ways off.
- techblueberry 1y ago
  I mean 20 years ago it was intelligence that could be used in multiple domains; the ability to reason in natural language. Which is what we have? Really, beyond this fantastical; AGI is when we have luxury automated space communism, I sort of legitimately don’t understand why ChatGPT 5 isn’t AGI. (Other than the fact that it would be super disappointing, which is maybe my point) Maybe it’s AGIv1? Maybe it’s AGI with an IQ of 47? But it’s super bipolar since it can also talk like someone with an IQ of 150x
  - acdha 1y ago
    Lack of reasoning or true understanding. It’s not just that it’s like IQ 47 but that it’s unreliable and inconsistent so you can’t safely deploy it in adversarial contexts.
    - techblueberry 1y ago
      Humans with true intelligence are unreliable and inconsistent, why would AGI be different?
      - acdha 1y ago
        AGI wouldn’t be, LLMs are because they’re still far from that level.
  - loeg 1y ago
    20 years ago, sub-human levels were extremely optimistic. The context has changed.
skywhopper 1y ago
This is Wikipedia’s definition “[AGI] is a type of artificial intelligence that would match or surpass human capabilities across virtually all cognitive tasks.” GPT-5 is nowhere close to this. What are you talking about?
- techblueberry 1y ago
  ChatGPT wrote this, but it is basically the argument I’m making: 1. Functional Definition of AGI If AGI is defined functionally — as a system that can perform most cognitive tasks a human can, across diverse domains, without retraining — then GPT-4/5 arguably qualifies: It can write code, poetry, academic papers, and legal briefs. It can reason through complex problems, explain them, and even teach new skills. It can adapt to new domains using only language (without retraining), which is analogous to human learning via reading or conversation. In this view, GPT-5 isn’t just a language model — it’s a general cognitive engine expressed through text. Again I think the common argument is more a religious argument than a practical one. Yes I acknowledge this doesn’t meet the frontier definition of AGI, but that’s because it would be sad if it was the case, not because there’s any actual practical sense that we’ll get to the sci-fi definition. This view that ChatGPT is already performing most tasks reasonably at the edge of beyond human ability is true.
  - tim333 1y ago
    People have different takes but the economically important point is when you can have AI do the jobs rather than having to hire humans. We are not there yet. GPT-5 is good at some things but not others.
    - techblueberry 1y ago
      That’s a good goal, but why is that “AGI”? Why is AGI a socio-political-economic metric and not a technical one, and if it is a socio-political-economic metric, than is just fantasy? Why are we spending trillions of dollars on something we can’t define in technical terms?
JKCalhoun 1y ago
I agree, we've been moving the goal posts. I think that the Turing Test was the first casualty of LLM ascendency. But I also think it's natural to move the goal posts. We try to peer at the future and what would convince us of machine intelligence. Academia finally delivers and we have to revise what we mean by intelligence. If one, settling a pillow by her head, Should say: "That is not what I meant at all; That is not it, at all."

junon 1y ago

Sam is wrong here, AGI has never had a clear definition, this there's no "we have" or "we haven't". I'd say most agree we're still a ways off.

techblueberry 1y ago
I mean 20 years ago it was intelligence that could be used in multiple domains; the ability to reason in natural language. Which is what we have? Really, beyond this fantastical; AGI is when we have luxury automated space communism, I sort of legitimately don’t understand why ChatGPT 5 isn’t AGI. (Other than the fact that it would be super disappointing, which is maybe my point) Maybe it’s AGIv1? Maybe it’s AGI with an IQ of 47? But it’s super bipolar since it can also talk like someone with an IQ of 150x
- acdha 1y ago
  Lack of reasoning or true understanding. It’s not just that it’s like IQ 47 but that it’s unreliable and inconsistent so you can’t safely deploy it in adversarial contexts.
  - techblueberry 1y ago
    Humans with true intelligence are unreliable and inconsistent, why would AGI be different?
    - acdha 1y ago
      AGI wouldn’t be, LLMs are because they’re still far from that level.
- loeg 1y ago
  20 years ago, sub-human levels were extremely optimistic. The context has changed.

skywhopper 1y ago

This is Wikipedia’s definition “[AGI] is a type of artificial intelligence that would match or surpass human capabilities across virtually all cognitive tasks.” GPT-5 is nowhere close to this. What are you talking about?

techblueberry 1y ago
ChatGPT wrote this, but it is basically the argument I’m making: 1. Functional Definition of AGI If AGI is defined functionally — as a system that can perform most cognitive tasks a human can, across diverse domains, without retraining — then GPT-4/5 arguably qualifies: It can write code, poetry, academic papers, and legal briefs. It can reason through complex problems, explain them, and even teach new skills. It can adapt to new domains using only language (without retraining), which is analogous to human learning via reading or conversation. In this view, GPT-5 isn’t just a language model — it’s a general cognitive engine expressed through text. Again I think the common argument is more a religious argument than a practical one. Yes I acknowledge this doesn’t meet the frontier definition of AGI, but that’s because it would be sad if it was the case, not because there’s any actual practical sense that we’ll get to the sci-fi definition. This view that ChatGPT is already performing most tasks reasonably at the edge of beyond human ability is true.
- tim333 1y ago
  People have different takes but the economically important point is when you can have AI do the jobs rather than having to hire humans. We are not there yet. GPT-5 is good at some things but not others.
  - techblueberry 1y ago
    That’s a good goal, but why is that “AGI”? Why is AGI a socio-political-economic metric and not a technical one, and if it is a socio-political-economic metric, than is just fantasy? Why are we spending trillions of dollars on something we can’t define in technical terms?

JKCalhoun 1y ago

I agree, we've been moving the goal posts. I think that the Turing Test was the first casualty of LLM ascendency. But I also think it's natural to move the goal posts. We try to peer at the future and what would convince us of machine intelligence. Academia finally delivers and we have to revise what we mean by intelligence. If one, settling a pillow by her head, Should say: "That is not what I meant at all; That is not it, at all."