The goalposts have not moved. The goalposts have never been moved. An AGI is an AI that can do everything a human can do, period. If you were starting a startup for example, you wouldn’t need to hire any humans - you would just spin up enough AGI instances and they would design your product, write your code, deploy it, handle your financials, respond to any and all customer interactions, proactively navigate regulations and litigation, and everything else that needs to be done in the management of a business. That is the goalpost for AGI. It’s an artificial human - a human replacement.

This is a popular take, but does it hold up to reality? From what I’ve seen most people have long expected AI to solve standardized tests, even more free form ones like the LSAT. LLMs’ new abilities are mostly just because of faster and cheaper training and huge amounts of data, but I don’t see anything it can solve that doesn’t use pattern matching.

There are many things that pattern matching over large amounts of data can solve, like eventually we can probably get fully generated movies, music compositions, and novels, but the problem is that all of the content of those works will have to have been formalized into rules before it is produced, since computers can only work with formalized data. None of those productions will ever have an original thought, and I think that’s why GPT-3’s fiction feels so shallow.

So it boils down to a philosophical question, can human thought be formalized and written in rules? If it can, no human ever has an original thought either, and it’s a moot point.

I think there are two different things that people are talking about when they say AGI - usefulness and actual general intelligence. I think we're already passed the point where these AIs are very useful and not just in a Siri or Google Assistant way and the goal posts for that have moved a little bit (mostly around practicality so the tools are in everyone's hands). But general intelligence is a much loftier goal and I think that we're eventually going to hit another road block regardless of how much progress we can make towards that end.

The goalposts never moved, but you're right that we're catching up quickly.

We always thought that if AI can do X then it can do Y and Z. It keeps turning out that you can actually get really good at doing X without being able to do Y and Z, so it looks like we're moving the goalposts, when we're really just realizing that X wasn't as informative as we expected. The issue is that we can't concretely define Y and Z, so we keep pointing at the wrong X.

But all indication is that we're getting closer.

We seem to be taking stands on either side of

> “there are/are not, additional properties to human level symbol manipulation, beyond what GPT encapsulates.”

GPT does appear to do an awful lot, before we find the limits, of pattern extrapolation.

No one has moved the goal posts. Let's see a computer pass a rigorous Turing test conducted by an interdisciplinary panel of expert evaluators. That has long been considered the gold standard for identifying the arrival of true AGI. GPT-4 is a tremendous technical achievement, but still far from that level.

The notion of some sort of technological "singularity" is just silly. It is essentially an article of faith, a secular religion among certain pseudo-intellectual members of the chattering class. There is no hard scientific backing for it.

The goalposts don't matter. If we all agreed today that we have AGI, nothing would be different tomorrow.

> We are moving the goal posts on AGI

What, in your mind, should the goal posts be for AGI?

I guess till some model explicitly says that it's sentient without any input, we would keep pushing the goal posts.