Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

DeepMind: A Generalist Agent

deepmind.com

Post reply on HN
Print view View original

91–100 of 345 posts

Re: DeepMind: A Generalist Agent

#91

Post by nmca »

PaulHoule wrote: post #49
axg11 wrote: post #22

Slowly but surely we're moving towards general AI. There is a marked split across general society and even ML/AI specialists between those who think that we can achieve AGI using current methods and those who dismiss the possibility. This has always been the case, but what is remarkable about today's environment is that researchers keep making progress contrary to the doubter's predictions. Each time this happens, th…

I have been impressed with what I've seen in the last six months but it still seems that GPT-3 and similar language models greatest talent is fooling people. The other day I prompted a language model with "The S-300 missile system is" and got something that was grammatical but mostly wrong: the S-300 missile system was not only capable of shooting down aircraft and missiles (which it is), but it was also good for sho…

Do you really, truly believe this problem is impossible to solve though? Even simple things make strides, eg: https://www.deepmind.com/publications/gophercite-teaching-la...

Re: DeepMind: A Generalist Agent

#92

Post by visarga »

2bitencryption wrote: post #10

given that the same model can both: 1. tell me about a cat (given a prompt such as "describe a cat to me") 2. recognize a cat in a photo, and describe the cat in the photo does the model understand that a cat that it sees in an image is related to a cat that it can describe in natural language? As in, are these two tasks (captioning an image and replying to a natural language prompt) so distinct that a "cat" in an im…

Check out "Flamingo"

https://twitter.com/serkancabi/status/1519697912879538177/ph...

Re: DeepMind: A Generalist Agent

#93

Post by TaupeRanger »

adamsmith143 wrote: post #90
dekhn wrote: post #47

Earlier quoted context omitted.

we have been using ML to solve useful problems in biology for more than 3 decades. However, it was usually called "advanced statistics and probability on large data sets" because, to be honest, that's what most modern ML is.

Of course there's no evidence that this isn't just what Human Brains are doing either.

There it is. The person who think human minds are python programs doing linear algebra.

Re: DeepMind: A Generalist Agent

#94

Post by PaulHoule »

dougabug wrote: post #83
PaulHoule wrote: post #49

Earlier quoted context omitted.

I have been impressed with what I've seen in the last six months but it still seems that GPT-3 and similar language models greatest talent is fooling people. The other day I prompted a language model with "The S-300 missile system is" and got something that was grammatical but mostly wrong: the S-300 missile system was not only capable of shooting down aircraft and missiles (which it is), but it was also good for sho…

Sure, but GPT-3 was trained by self-supervised learning on only static text. We see how powerful even just adding captions to text can be with the example of DALLE-2. GATO takes this further by letting the large scale Transformer learn in both simulated and real interactive environments, giving it the kind of grounding that the earlier models lacked.

I will grant that the grounding is important.

The worst intellectual trend of the 20th century was the idea that language might give you some insight into behavior (Sapir–Whorf hypothesis, structuralism, post-structuralism, ...) whereas language is really like the evidence left after a crime.

For instance, language maximalists see mental models as a fulcrum point for behavior, and they are, but they have nothing to do with language.

I have two birds that come to my window. One of them has no idea of what the window is and attacks her own reflection hundreds of times a day. She can afford to do it because her nest is right near the bird feeder and doesn't need to work to eat, in fact it probably seems meaningful to her that another bird is after her nest. This female cardinal flies away if I am in the room where she is banging.

There is a rose-breasted grosbeak, on the other hand, that comes to the same window. She doesn't mind if I come close to the window, instead I see her catch the eye of her reflection and then catch my eye. She basically understands the window.

Here you have two animals with two different acquired mental models... But no language.

What I like about the language-image models is how the image grounds reality outside language, and that's important because the "language instinct" is really a peripheral that attaches to an animal brain. Without the rest of the animal it's useless.

Re: DeepMind: A Generalist Agent

#95

Post by browningstreet »

gitfan86 wrote: post #86
axg11 wrote: post #22

Slowly but surely we're moving towards general AI. There is a marked split across general society and even ML/AI specialists between those who think that we can achieve AGI using current methods and those who dismiss the possibility. This has always been the case, but what is remarkable about today's environment is that researchers keep making progress contrary to the doubter's predictions. Each time this happens, th…

Tesla FSD is quickly becoming less of a software problem and more of a problem of semantics. If the car drives someone to and from work 30 days in a row without a problem, is it truly FSD? What about 300 days? Where do you draw the line? 1000x safer than the average human? Same thing here will AI. How many conversations with GTP-X need to happen without a stupid response from GTP before we call it real world AI?

Do we account for stupid responses from humans in human communication in the targets?

Re: DeepMind: A Generalist Agent

#96

Post by chilmers »

password54321 wrote: post #82
chilmers wrote: post #8

This sounds exciting, but the example outputs look quite bad. E.g. from the interactive conversation sample: > What is the capital of France? > Marseille And many of the generated image captions are inaccurate.

Yeah they put that example for a reason. Read the paper and stop acting like this is some great insight that you discovered.

What exactly did I say that implied I was acting as this was a “great insight I’d discovered”? That’s a rather rude and unfair insult I’d say.

Re: DeepMind: A Generalist Agent

#97

Post by TaupeRanger »

axg11 wrote: post #22

Slowly but surely we're moving towards general AI. There is a marked split across general society and even ML/AI specialists between those who think that we can achieve AGI using current methods and those who dismiss the possibility. This has always been the case, but what is remarkable about today's environment is that researchers keep making progress contrary to the doubter's predictions. Each time this happens, th…

And yet, the only thing that really matters out of your entire list is the 1st one: that AI solves problems that actually improve the human condition. And Alpha Fold has not done that at all. It may be very nice for people interested in protein folding, but until it actually helps us find something that we wouldn't have found otherwise, and that discovery leads to (for example) an ACTUAL drug or treatment that helps real patients, AND that drug/treatment is actually BETTER than what is already available by helping people live longer or better lives, AI has done nothing. In effect, AI has STILL done nothing meaningful. One could argue, through the use of predatory algorithms, that the ONLY thing it has done is harm.

Re: DeepMind: A Generalist Agent

#98

Post by PaulHoule »

nmca wrote: post #91
PaulHoule wrote: post #49

Earlier quoted context omitted.

I have been impressed with what I've seen in the last six months but it still seems that GPT-3 and similar language models greatest talent is fooling people. The other day I prompted a language model with "The S-300 missile system is" and got something that was grammatical but mostly wrong: the S-300 missile system was not only capable of shooting down aircraft and missiles (which it is), but it was also good for sho…

Do you really, truly believe this problem is impossible to solve though? Even simple things make strides, eg: https://www.deepmind.com/publications/gophercite-teaching-la...

If you've been involved in efforts to develop advanced technologies you might eventually encounter an

https://en.wikipedia.org/wiki/Asymptote

which is described as a risk in great detail

https://www.amazon.com/Friends-High-Places-W-Livingston/dp/0...

it's quite a terrible risk because you often think "if only I double or triple the resources I apply to do this I'll get it." Really though you get from 90% there to 91% to 92% there.... You never get there because there is a structural mismatch between the problem you have and how you're trying to solve it.

My take is that people have been too incredulous about the idea that you can just add more neurons and train harder and solve all problems... But if you get into the trenches and ask "why can't this network solve this particular task?" you usually do find structural mismatches.

What's been exciting just recently (last month or so) are structurally improved models which do make progress beyond the asymptote because they are confronting

https://www.businessballs.com/strategy-innovation/ashbys-law...

Re: DeepMind: A Generalist Agent

#99

Post by svieira »

dekhn wrote: post #45
ReadEvalPost wrote: post #38

Earlier quoted context omitted.

Certainly we can say our ML models are becoming more general in the sense of being able to cross-correlate between multiple domains. This is quite a different story than "becoming a general intelligence." Intelligence is a property of a being with will. These models, and machines in general, do not posses will. It is we who define their form, their dataset, their loss function, etc. There is no self-generation that m…

Assumption of will is unfounded, scientifically speaking. Your entire argument is philosophical, not scientific. The subjective experience of free will is in no way unrefutable proof that will is required for intelligence.

Since a working (in the sense of 'working title') ontology and epistemology are _required_ for science (read "natural philosophy") is this argument not arguing that "the argument for quarks is unfounded, biologically speaking"? That said, I _believe_ that both Aristotle and St. Thomas agree with you that will and intellect are not necessarily connected, so you could have an intellectual power with no freedom to choose.

Re: DeepMind: A Generalist Agent

#100

Post by phyalow »

Isnt this a general reinforcement learning agent with a transformer as the policy discriminator? Very cool, but not necessarily a giant leap forward, more like a novel combination of existing tools and architectures. Either way impressive.
Post reply on HN