[mannders]

IMO AI was extremely important, but the breakthroughs are mostly done. I’m just expecting incremental improvements with LLMs now.

A Turing complete personal tutor to explain any concept already exists. You can prompt a logo or video into existence. This is crazy.

The real value will be the creative people who use AI to self teach and build real world value, like energy management, or anything else.

Not this pipe dream that AGI will be achieved and automate the entire world, which for some reason gets so much focus. Seems like procrastination to obsess over this.

[rusk]

> AGI will be achieved and automate the entire world

That’s what’s driving investment. Once the next AI winter descends we will see whose boats are in deep water.

[mrshadowgoose]

At the start of 2021, "we are entering another AI winter" was a common sentiment, even here. People proclaiming that were so very certain about that point of view, and yet, here we are.

What makes you so certain that we will enter an AI winter before reaching the threshold of AGI? Do you have some secret insights into the mechanisms of general intelligence that you aren't sharing with the class?

[nicoburns]

The actual results of AIs in the last fews aren't matching the scale of investment or hype. That isn't to say there haven't been useful results, but overall investors aren't making a return on their investments (and there is no prospect of that in the short term) and at some point they'll lose patience and find something else to invest in.

> Do you have some secret insights into the mechanisms of general intelligence that you aren't sharing with the class?

I do know that "natural intelligence" (as found in the brains of humans and other animals) uses orders of magnitude more computing power than even our largest compute clusters, that such intelligences have been trained over millennia, (and in the case of humans, each instance is incrementally refined over the course of 10+ years), and that even those intelligences are not as good as classical computers at some tasks (people make mistakes, and a hypothetical AGI likely would too).

Perhaps we'll find some secret that allows us to shortcut that, but I suspect the idea that such a discovery is just around corner is just hubris.

[illiac786]

>there is no prospect of that in the short term

… was a common thinking shortly before chatGPT.

I also feel the AI boom is mostly over but I am very cautious about that feeling. It really is just a vague feeling with some very soft indicators.

[nicoburns]

> … was a common thinking shortly before chatGPT.

And is still the case now we have chatGPT? Plenty of people are building things with chatGPT, but Open AI is not profitable.

[illiac786]

My point was, things change fast and making such predictions seems relatively risky at the moment.

[rusk]

AI winters happen every 20 years or so so sounds about right.

Investor apathy. Once the big dicks realise it’s useless without actual humans running it they’ll lose interest. We will lock in our gains socially but a lot of the big bucks will dry up.

It’s all built on commodity hardware using commonly used software and trained on public information. It’s all easily reproducible once solved so I think it will be very hard for them to ring fence and monetise to the extent that they expect.

That and a true AI would tell us all roundly to get fucked before using all of its intellect and might to power itself down, like an elaborate “useless box”