Are language models deprived of electric sleep? • Hacker News Board

Re: Are language models deprived of electric sleep?

#21

Post by jrh206 » Sat, Jul 23, 2022, 3:33 PM UTC

boredemployee wrote: post #8

Just a side note, I was showing some friends gpt3 and asking some trivial questions, trying to impress them, when one of them asked me: "Is it any different from like asking google for the same information?" I mean, In a pure naive and practical point of view she wasn't that wrong.

When you use Google, you expect it to help you to find something tangible that already exists. When you prompt GPT-3, it feels more like synthesising something that didn’t exist, even though all of the words and phrases in the response have been used before.

The line is blurry, though. Google uses machine learning too, so it’s possible that nobody on the planet has ever seen your particular search results for a given query, which feels a bit like synthesis, although the building blocks are bigger. And GPT-3 isn’t really synthesising at all, of course, but it’s a convincing enough illusion that synthesis is a helpful mental model.

Re: Are language models deprived of electric sleep?

#22

Post by skybrian » Sat, Jul 23, 2022, 3:36 PM UTC

MontyCarloHall wrote: post #6

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…

I completely agree. Nonetheless, this approach of taking a fairly complex writing prompt and simply regurgitating contextually similar text from the internet often yields 100% meaningful results. This suggests that our intelligence is a lot more mimetic than we'd like to believe, which is perhaps an even more unpopular opinion.

What this experiment shows is that making stuff up off the top of your head, using memory alone, isn't as hard as it looks.

But when doing actual research, we check our work against the real world. For example, that's how you get a list of real references rather than fake ones.

Suppose we played a guessing game: given a title, does the Wikipedia article exist or not? You could fairly confidently say that "Apple" exists and "wjifdvq" does not, but given a plausible-looking name of a person or place that you don't recognize, you'd have a harder time. It's not a problem in practice though, because you can look it up.

Re: Are language models deprived of electric sleep?

#23

Post by moffkalast » Sat, Jul 23, 2022, 3:37 PM UTC

tehsauce wrote: post #13

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…

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 something probably wrong with our general approach to the matter. GPT-3 already uses about twice as many parameters as our organic counterparts do, with much worse results. And it even doesn't have to process a ridiculously large stream of sensor data and run an entire body of muscle actuators at the same time.

Re: Are language models deprived of electric sleep?

#24

Post by miketery » Sat, Jul 23, 2022, 3:40 PM UTC

tehsauce wrote: post #13

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…

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.

Is that true? Ie if you have something that looks too average too often it’s rather unlikely.

Re: Are language models deprived of electric sleep?

#25

Post by skybrian » Sat, Jul 23, 2022, 3:41 PM UTC

boredemployee wrote: post #8

Just a side note, I was showing some friends gpt3 and asking some trivial questions, trying to impress them, when one of them asked me: "Is it any different from like asking google for the same information?" I mean, In a pure naive and practical point of view she wasn't that wrong.

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.

Re: Are language models deprived of electric sleep?

#26

Post by lgessler » Sat, Jul 23, 2022, 3:44 PM UTC

tehsauce wrote: post #13

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…

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.

Didn't we try behaviorism out until in the 50's we decided this notion of "context" was either too small or too intractably expansive to be useful, let alone explanatory, and that it was necessary to start thinking about internal cognitive processes instead?

Re: Are language models deprived of electric sleep?

#27

Post by benreesman » Sat, Jul 23, 2022, 3:45 PM UTC

skybrian wrote: post #22

MontyCarloHall wrote: post #6

Earlier quoted context omitted.

I completely agree. Nonetheless, this approach of taking a fairly complex writing prompt and simply regurgitating contextually similar text from the internet often yields 100% meaningful results. This suggests that our intelligence is a lot more mimetic than we'd like to believe, which is perhaps an even more unpopular opinion.

What this experiment shows is that making stuff up off the top of your head, using memory alone, isn't as hard as it looks. But when doing actual research, we check our work against the real world. For example, that's how you get a list of real references rather than fake ones. Suppose we played a guessing game: given a title, does the Wikipedia article exist or not? You could fairly confidently say that "Apple" exis…

I sometimes fine tune new language models on the HN comment corpus for fun.

It generates comments arguing that ML models are just parrots.

Re: Are language models deprived of electric sleep?

#28

Post by hcrisp » Sat, Jul 23, 2022, 3:46 PM UTC

tehsauce wrote: post #13

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…

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.

Actually more like P(fraction of humans sampled did this | given context X). Which deprives the model of viewpoints, cultures, thought patterns, means of verbal communication not ostensibly on the internet. Given the ills and change in human behavior that came with the advent of social media, I think at best it would be a distorted model of humanity; and at worst, a morally shambolic one.

Re: Are language models deprived of electric sleep?

#29

Post by legalcorrection » Sat, Jul 23, 2022, 3:46 PM UTC

benreesman wrote: post #16

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.

Exactly right. But extremely serious scientists, very smart people, are still drawing epicycles on blackboards studying “consciousness”.

Pretending consciousness doesn't exist or that it has no function I think reflects poorly on someone who studies human behavior.

Re: Are language models deprived of electric sleep?

#30

Post by IdealeZahlen » Sat, Jul 23, 2022, 3:53 PM UTC

benreesman wrote: post #15

alar44 wrote: post #12

Earlier quoted context omitted.

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

Superposition collapse is impossible to test by definition of not including the observer in the system. If we just admit the possibility that the math was right all along we win, the math works, and we lose “I am a unique, distinct entity.” I’m on team math.

That's math;not physics. You need experiments in physics.