>Altmans quote was that AGI "could be just a few thousand days away" which sounds a lot more optimistic on the surface than it actually is.

I think he was referring to ASI, not AGI.

Isn't ASI > AGI?

Both are poorly defined.

By all the standards I had growing up, ChatGPT is already AGI. It's almost certainly not as economically transformative as it needs to be to meet OpenAI's stated definition.

OTOH that may be due to limited availability rather than limited quality: if all the 20 USD/month for Plus gets spent on electricity to run the servers, at $0.10/kWh, that's about 274 W average consumption. Scaled up to the world population, that's approximately the entire global electricity supply. Which is kinda why there's also all the stories about AI data centres getting dedicated power plants.

Don't know why you're being downvoted, these models meet the definition of AGI. It just looks different than perhaps we expected.

We made a thing that exhibits the emergent property of intelligence. A level of intelligence that trades blows with humans. The fact that our brains do lots of other things to make us into self-contained autonomous beings is cool and maybe answers some questions about what being sentient means but memory and self-learning aren't the same thing as intelligence.

I think it's cool that we got there before simulating an already existing brain and that intelligence can exist separate from consciousness.

Is the S here referring to Sentient or Specialised?

Scottish.

Super(human).

Old-school AI was already specialised. Nobody can agree what "sentient" is, and if sentience includes a capacity to feel emotions/qualia etc. then we'd only willingly choose that over non-sentient for brain uploading not "mere" assistants.

Super, whatever that means

Actually, the S means hope.

Given that ChatGPT is already smarter and faster than humans in many different metrics. Once the other metrics catch up with humans it will still be better than humans in the existing metrics. Therefore there will be no AGI, only ASI.

My fridge is already smarter and faster than humans in many different metrics.

Has been this way since calculation machines were invented hundreds of years ago.

_Thousands_; an abacus can outperform any unaided human at certain tasks.

Between 1 and 10 thousands of days, so 3 to 27 years.

A range I'd agree with; for me, "pessimism" is the shortest part of that range, but even then you have to be very confident the specific metaphorical horse you're betting on is going to be both victorious in its own right and not, because there's no suitable existing metaphor, secretly an ICBM wearing a patomime costume.

1 you use 1

2 (or even 3) you use "a couple"

A few is almost always > 3 and one could argue that upper limit 15

So, 10 years to 50 years

Personally speaking, above 10 thousand I'd switch to saying "a few tens of thousands".

But the mere fact you say 15 is arguable does indeed broaden the range, just as me saying 1 broadens it in the opposite extent.

You imply that he knows exactly when which imo is not and could even be next year for what we knows.. Who know every paper yet to be published??

few is not > 3. Literally it's just >= 2, though I think >= 3 is the common definition.

15 is too high to be a "few" except in contexts of a few out of tens of thousands of items.

Realistically I interpret this as 3-7 thousands of days (8 to 19 years), which is largely consensus prediction range anyway.

While it's not really _wrong_ to describe two things as 'a few', as such, it's unusual and people don't really do it in standard English.

That said, I think people are possibly overanalysing this very vague barely-even-a-claim just a little. Realistically, when a tech company makes a vague claim about what'll happen in 10 years, that should be given precisely zero weight; based on historical precedent you might as well ask a magic 8-ball.

Just in time for them to figure out fusion to power all the GPUs.

But really. o1 has been very whelming, nothing like the step up from 3.5 to 4. Still prefer sonnet3.5 and opus.

OpenAI is a Microsoft play to get into power generation business, specifically nuclear, which is a pet interest of Bill Gates for many years.

There, that's my conspiracy theory quota for 2024 in one comment.

I don't think Gates has much influence on Microsoft these days.

He controls approximately 1% of the voting shares of MSFT.

And I would argue his "soft power" is greatly diminished as well

It's kinda cool as a conspiracy theory. It's just reasonable enough if you don't know any of the specifics. And the incentives mostly make sense, if you don't look too closely.

> it's possible that we will have superintelligence in a few thousand days

Sure, a few thousand days and a few trillion $ away. We'll also have full self driving next month. This is just like the fusion is the energy of the future joke: it's 30 years away and it will always be.

Now it’s 20 years away! It took 50 years for it to go from 30 to 20 years away. So maybe, in another 50 years it will be 10 years away?

Because as we all know: Full Self Driving is just six months away.

Thanks, now I cannot unthink of this vision: developers activate the first ASI, and after 3 minutes it spits out full code and plans for a working Full Self Driving car prototype:)

I thought super-intelligence was to say self driving would be fully operational next year for 10 consecutive years?

My point was that only super intelligence could possibly solve a problem that we can only pretend to have solved.

To paraphrase a notable example: We will have full self driving capability next year..