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

Are language models deprived of electric sleep?

blog.cbs.dk

Post reply on HN
Print view View original

51–60 of 95 posts

Re: Are language models deprived of electric sleep?

#51

Post by Tao3300 »

boredemployee wrote: post #34
skybrian wrote: post #25

Earlier quoted context omitted.

Yes, of course it's different. Google can point you to a real Wikipedia article while GPT-3 will generate links to fake ones (or you could ask it to generate a fake article.) It's the difference between truth and fiction. Or at least, a first-level approximation. Wikipedia articles can be wrong.

Well google also can point me to fake links with no real content at all as well. But I get your point.

Google can also return "People also asked" results that make so sense or give incorrect answers. Sometimes they even contradict the cited article.

E.g

Ask "is pi disjunctive" and you'll get the top result of

> = 3.1415926535897932384626433832795... is a disjunctive number.

But if you click into the article (what percentage even will?) you'll find it's part of a larger statement:

> It is not known whether π = 3.1415926535897932384626433832795... is a disjunctive number.

Re: Are language models deprived of electric sleep?

#52

Post by anonymoushn »

Tao3300 wrote: post #51
boredemployee wrote: post #34

Earlier quoted context omitted.

Well google also can point me to fake links with no real content at all as well. But I get your point.

Google can also return "People also asked" results that make so sense or give incorrect answers. Sometimes they even contradict the cited article. E.g Ask "is pi disjunctive" and you'll get the top result of > = 3.1415926535897932384626433832795... is a disjunctive number. But if you click into the article (what percentage even will?) you'll find it's part of a larger statement: > It is not known whether π = 3.141592…

Earlier this year you could google "is covid airborne" and get a WHO article explaining that it is not airborne above the first result :)

Re: Are language models deprived of electric sleep?

#53

Post by soraki_soladead »

moffkalast wrote: post #23
tehsauce wrote: post #13

Earlier quoted context omitted.

Worth noting that in the limit the distribution P(human did/said A | context B) is a complete definition of human behavior. If you could model this perfectly, that would be a perfect model of a human ie identical.

That's the brute force approach anyway. Even if you do take an inordinate amount of data to sample from you'll likely get something that's woefully impractical to operate, even if it produces vaguely human like responses. We do on the other hand know for a fact it's possible to run an instance of consciousness in a volume of about a liter that consumes like 20 watts (aka your average human brain), so there's somethin…

> GPT-3 already uses about twice as many parameters

This isn't accurate. GPT-3 has 175B parameters. The human brain has ~175B cells (neurons, glia, etc.) The analog to GPT-3's parameter count would be synapses, not neurons, where even conservative estimates put the human brain at several orders of magnitude larger. It's likely that >90% of the 175B could be pruned with little change in performance. That changes the synapse ratios since we know the brain is quite a bit sparser. In addition, the training dataset is likely broader than the majority of Internet users. Basically, its not an apples-to-apples comparison.

That said, I agree that simply scaling model and data is the naive approach.

Re: Are language models deprived of electric sleep?

#54

Post by leobg »

benreesman wrote: post #2

I’m well aware that this is an unpopular opinion. When your game plan is take a multi-head attention transformer and overfit it on as much of the Internet as we have A100s to fit it on, you’re going to get a distribution, and if you sample from it, it’s going to sound like the Internet. People read too much into this. The technology is obviously powerful borderline dangerous, but it’s still just maximizing P(this-sen…

There’s a case to be made for the proposition that diffusion is what makes the kind of “intelligence” that we value. If you cook a great dish and you cannot explain how you did it, you get a lot of credit for being a great cook. If, on the other hand, you can point to exactly one recipe that you followed, you will get a lot less credit. So in a sense, the harder it is to trace down the inspirations from which a composition sprung, the higher we value it.

Re: Are language models deprived of electric sleep?

#55

Post by benreesman »

omnicognate wrote: post #46
alar44 wrote: post #12

Earlier quoted context omitted.

No, they're "dragging their feet" because it's impossible to test, it's unfalsifiable.

And because if every outcome happens the probabilities of those outcomes are meaningless, and those probabilities are the predictive content of a quantum mechanical theory. It's an "interpretation" that requires you to ditch the entire value of the thing you're "interpreting". You don't need to know anything about QM other than that it has been empirically tested and is probabilistic in nature to reject the many worl…

Hey if you’ve got David Deutsch dead to rights you should book a meeting and have him hand you his Fellowship in the Royal Society.

I’m not a physicist but I am an interested layman, I’m pretty sure I’ll catch wind of the Nobel Prize you’d win for proving him wrong about everything.

Re: Are language models deprived of electric sleep?

#56

Post by benreesman »

Swizec wrote: post #40
benreesman wrote: post #27

Earlier quoted context omitted.

I sometimes fine tune new language models on the HN comment corpus for fun. It generates comments arguing that ML models are just parrots.

As a bird owner, they are not “just parrots”. There is clear intent to communicate. Even species who can’t say human words use distinct vocalizations for specific purposes and situations.

I didn’t say that fine-tuned language models have enlightened opinions about living things, just that they produce comments indistinguishable to me from the GP :)

Re: Are language models deprived of electric sleep?

#57

Post by benreesman »

legalcorrection wrote: post #35
benreesman wrote: post #31

Earlier quoted context omitted.

I’m saying that the idea of a bright line between the emergent behavior of a dolphin and a human is very pre-Copernicus. Studying, even measuring the capabilities of an animal is science. Justifying a soul is the purview of spirituality, not science. (Nothing against spirituality, I have a spiritual life, I just don’t confuse it with science).

I strongly disagree. Pure behaviorism is just willful blindness. Consciousness is a real phenomenon, as any person not philosophically committed to denying its existence can tell you. It's front and center of our experience of human cognition. It would be quite strange for it to serve no function in the human mind. Yeah, it's hard to quantify and isolate and experiment on, but that just speaks to either current limit…

You strongly disagree that other intelligent, social, creative animals are built along similar lines to Homo sapiens?

You really think that we’re a special case, that a difference in degree has become a difference in kind?

I personally experience a feeling that I’m conscious subjectively, but I have no evidence that I’m any more or less motivated by pleasure or pain or community than a dolphin is.

Where do we draw the line? What’s the acid test for “yup now we’re dealing with consciousness”?

Re: Are language models deprived of electric sleep?

#58

Post by benreesman »

Veedrac wrote: post #42
benreesman wrote: post #2

I’m well aware that this is an unpopular opinion. When your game plan is take a multi-head attention transformer and overfit it on as much of the Internet as we have A100s to fit it on, you’re going to get a distribution, and if you sample from it, it’s going to sound like the Internet. People read too much into this. The technology is obviously powerful borderline dangerous, but it’s still just maximizing P(this-sen…

1. They aren't overfit. 2. It was far from obvious a priori that this could work at all. The existence of the sentiment ‘lol you trained a big model on X so obviously it produces a good probability distribution on X’ only exists because big models proved to be extraordinarily more effective than anyone expected.

When you’re churning through terrawords of training data and aren’t cross-validating against proportionally-sized hold out sets?

It’s not a certainty that you’re overfitting but the burden is to show otherwise.

Besides, if your corpus is asymptomatically everything, why wouldn’t fitting it perfectly fitting it be the goal?

It would probably be more accurate to say that the bias/variance tradeoff loses meaning as the training set goes to infinity.

Grapevine is that the big actors are going to 5x their flops at any price in the next three years.

What does validation loss even mean at that scale?

Re: Are language models deprived of electric sleep?

#59

Post by benreesman »

the8472 wrote: post #48
benreesman wrote: post #2

I’m well aware that this is an unpopular opinion. When your game plan is take a multi-head attention transformer and overfit it on as much of the Internet as we have A100s to fit it on, you’re going to get a distribution, and if you sample from it, it’s going to sound like the Internet. People read too much into this. The technology is obviously powerful borderline dangerous, but it’s still just maximizing P(this-sen…

Part of the evaluation procedures for the models like these is checking whether the output they produce is present in the training set. The fraction of samples being verbatim reurgitations is low unless you prompt-engineer your way to that, copilot being able to recite the fast inverse square root function being a common example. But the ability to recite some samples is not proof of inability to create novel outputs…

Oh sure. I didn’t mean to imply that GPT-6 will have memorized Wikipedia.

As you start to get to the size and comprehensiveness in the corpus that’s going on now, novel model outputs approach being something that might not have been in the training set, but likely will be in the future.

Re: Are language models deprived of electric sleep?

#60

Post by kadonoishi »

moffkalast wrote: post #23
tehsauce wrote: post #13

Earlier quoted context omitted.

Worth noting that in the limit the distribution P(human did/said A | context B) is a complete definition of human behavior. If you could model this perfectly, that would be a perfect model of a human ie identical.

That's the brute force approach anyway. Even if you do take an inordinate amount of data to sample from you'll likely get something that's woefully impractical to operate, even if it produces vaguely human like responses. We do on the other hand know for a fact it's possible to run an instance of consciousness in a volume of about a liter that consumes like 20 watts (aka your average human brain), so there's somethin…

Is evolution not brute-force?
Post reply on HN