Our world is changing before our very eyes while this guy is belaboring the technicalities. You could hardly ask for a keener display of the philosophical gulf between scientists and engineers.
Especially since the rise of GPT-3 and now these image models, we've seen the pop-culture face of AI become even narrower. The promise of generalization that could lead to intelligent behavior has given way to people sharing amusing pictures or phrases that these models have generated, because that's what they do. It's cool, but it's basically become orthogonal to any AGI, or even AI with applications. It's now just a neat cultural phenomenon from which laypeople somehow extrapolate the kind if stuff the parent is saying.
I'm not saying AI (neural networks) isn't making research progress, it's just that it has almost nothing to do with any of what laypeople extrapolate from it
https://twitter.com/AdeptAILabs/status/1570144499187453952 https://twitter.com/runwayml/status/1568220303808991232
Text to video and converting some selected requests into actions is all nice, but it hardly contradicts the GP's observation: it's nowhere near AGI.
EDIT Since you ninja edited this in:
> Text to video and converting some selected requests into actions is all nice, but it hardly contradicts the GP's observation: it's nowhere near AGI.
If you review the root comment I made, you'll understand that I was never arguing with the GP about AGI in the first place.
At this point, I'm wondering if you're just provoking me deliberately. The comment I replied to said the following:
> The promise of generalization that could lead to intelligent behavior has given way to people sharing amusing pictures or phrases that these models have generated, because that's what they do. It's cool, but it's basically become orthogonal to any AGI, or even AI with applications.
And then I posted evidence of concrete applications that are in progress at some of the most well-resourced companies in Silicon Valley. Absolutely groundbreaking stuff that more than prove sophisticated applications of contemporary AI are well on their way to being realized.
A lot of "histrionic phrases" to describe your reading comprehension ability are occurring to me right now, but I'll refrain from using them.
A Twitter thread demo is not quite a revolution yet IMHO. Even in the 60s some people thought ELIZA was a real person.
I've said all I have to say. Have a nice day.
Neither of the 2 companies you posted marketing materials about is among "some of the most well-resourced companies in Silicon Valley".
Two of the founders of Adept AI are authors on the paper 'Attention Is All You Need'. If you don't understand the significance of that, then you're speaking well outside of what you're qualified to comment on. The company has also raised capital from top tier SV investors.
Runway ML has raised money from Lux Capital.
These companies are not just well-resourced, they are positioned in the upper echelon of the innovation business.
> It is easy to bet against new paradigms in their beginning stages: the Copernican heliocentric model of cosmology was originally less predictive of observed orbits than the intricate looping geocentric competitor. It is simple to play around with a large language model for a bit, watch it make some very discouraging errors, and throw in the towel on the LLM paradigm. But the inexorable scaling laws of deep learning models work in its favor. Language models become more intelligent like clockwork due to the tireless work of the brilliant AI researchers and engineers concentrated in a few Silicon Valley companies to make both the model and the dataset larger.
I don't know about you, but if I feed a program with hundreds of billions of "parameters" a huge chunk of the internet and it can then kinda-sorta do a bunch of things, sometimes semi-intelligently, but for the most part couldn't compete with a 4-year-old child... I'd say that's more on the Ptolemaic side of things than the Copernican side. Certainly "it gets better as you feed it more data" is equally true of both paradigms, so I'm not sure what Roon's point is here.
The appeal to the Copernican revolution itself has a bit of a hype-y, cranky odor. Virtually every crank appeals to Copernicus as a role model and vindicator. Real scientists usually don't, because they are busy with the hard, humbling business of actually figuring out how the world works.
Now don't get me wrong, I am thrilled by the research advances of the last couple decades, the foundation models, AlphaGo and AlphaFold, etc. The action model from Adept is great and Adept may become a very successful company. It's all very cool. But every paradigm shift in AI has been heralded as the thing that will Change Everything, and they usually don't. Big, exciting shifts in research don't necessarily mean as much in practice right away. I tend to think that getting AI "right enough" to have a huge, pervasively transformative impact on human life is going to take quite a few decades at least, if not centuries or more.
The recent progress on generative models is a major research achievement, to be sure. That said, I'm not sure what it means it "terraform society," but so far AI shows no signs of making the same magnitude of impact on society as, say, the S-tier technological advances of the 20th and early 21st centuries, such as the personal computer, Internet, smartphone, or atomic bomb. That all may change if we get AGI that actually works, of course.
I’m the opposite. I’m finally, after a long time, starting to get excited about AI. Yes, most outputs still suck and require a lot of experimentation and rephrasing, and yes, midjourney produces a lot of same-looking things (less freedom, but also less crap compared to dall-e).
But wow, now even I, someone with no artistic talent whatsoever, can with just a few prompts create a cool illustration. My current discord avatar is a sloth drinking a cocktail [0]. Zoomed in, it looks a bit uncanny, but generally and especially at smaller sizes, it’s fine.
I could not draw something even halfway as okay. I would not want to pay someone to do it for me as it’s of no big importance to me (I once paid someone for their Stranger Things as sloths image, but even that was just something they already created, not a commission which would have been vastly more expensive).
Personally, I really can’t wait what the next generation will be like, and what it will enable people to do, what they will enable me to do. Yes, I’m very excited.
https://twitter.com/AdeptAILabs/status/1570144499187453952 https://twitter.com/runwayml/status/1568220303808991232
Because I didn't say it will "obsolete" human labor.
You're right that the word "will" is strong in that statement. But, since basically nothing is absolute in a philosophical sense and certainly no one can prognosticate with certainty, it's fair to read that as "will with high likelihood".
And your surgical nitpicking doesn't hamper my argument in quite the way you think it does. Frankly, I believe it's in defiance of the following HN guideline:
> Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith.