12 Comments
User's avatar
Richard Jaffee's avatar

Cool summation and look into the future!

Philosophy bear's avatar

I don't think Marcus would see himself as supporting level 0

Dan's avatar

This article doesn't actually disambiguate between common conceptions of AGI, instead it's disambiguating between forecasts of plateaued LLM utility?

These are two very different things?

For all their faults I don't think Gary Marcus or Timnit Gebru think AGI is definitionally "a giant useless scam", they just think LLMs are.

Arnold v.E.'s avatar

The article lays out the different streams of thought regarding the ongoing revolution in AI technologies and what people believe the development of AGI means for the future. It is not trying to provide different commons defintions of AGI. See the title of the article, which alludes to the question ("How AGI-pilled are you?" or, in other words, "how convinced are you by the most extreme version of an AGI-driven future?")

David J Higgs's avatar

Given that previous attempts to narrowly/technically/intrinsically define AGI have tended to lead us astray w/ regards to predicting the future (e.g. the "Turing test"), I think this kind of capabilities/extrinsic definition makes more sense. It's also in good company w/ many similar concepts like TEDAI (top expert dominating AI), Transformative AI, "Drop in Remote Worker" AGI, Geniuses in a Data Center, etc.

Yitz's avatar
Aug 5Edited

The position ex-tech among the average American seems to be between 1-2. People still have a very hard time groking 3+ particularly because I think it does pose some pretty foundational questions to concepts that people accept as base reality (e.g. Market-Based Capitalism, having a 9-5, Democracy) and the Labs have still not produced good enough descriptions in detail about how they envision basic societal questions (e.g. In a radically abundant society who gets to own Malibu real estate, the last family that happened to acquire it and now it’s set forever?). 4-5 are even more outside the Overton norms.

blake harper's avatar

You had me at Wittgenstein. But these levels aren't MECE, and that's annoying ;P

You can, for example, believe LLMs are a fuzzy jpeg of the internet while still believing the technological impact will be somewhere between very useful SaaS and the internet. Most bankers and management consultants I know are in that camp, as am I.

I'd encourage you to go deeper on the Wittgenstein angle. Why don't you start by asking your favorite LLM what Wittgenstein would think about the very idea of AGI? That'll send you down a good rabbit hole on the foundations of cognitive science, its alienation from biology, and alternative ways of thinking about intelligence — especially deflationary, non-reductionist, and enactivist approaches.

Juliana's avatar

This is so good

fox's avatar
2dEdited

>This is the “AI as normal technology” view, and the view of much of the San Francisco tech ecosystem, especially “e/acc”3 and “builder” types. It would be roughly accurate if model capabilities plateaued around 2025-2026 levels.

I think this misunderstands the “AI as normal technology” view. AIANT assumes progress continues at its current rate. There would have to be a qualitative shift up an out of the page to overcome the real world mechanics that our model is based on.

Ha Tran Nguyen Phuong's avatar

Where do you think Ray Kurzweil is on this scale?

David J Higgs's avatar

Wonderful taxonomy. I am unsure whether I like this more or Zvi's simpler 3 (/4) level taxonomy of (not) AI-pilled, AGI-pilled and ASI-pilled.

Falling Rocket's avatar

This is a great piece. However, as other commenter noted you didn't actually disanbiguate 'AGI', but instead divided the people forecasting AI progress into 6 discrete tribes.