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

LaMDA is not sentient

garymarcus.substack.com

Post reply on HN
Print view View original

121–130 of 694 posts

Re: LaMDA is not sentient

#121

Post by mobiuscog »

sweezyjeezy wrote: post #109
tinco wrote: post #75

Earlier quoted context omitted.

I'm sorry but you completely fail at explaining what the difference is between GPT-3 and a human when generating sentences. And frankly I don't believe there is any significant difference. If you inform GPT-3 of what it has eaten then it can answer questions about it just as well as a human can, probably better. Why would you assume human thinking is different from generating the next N language tokens based on previ…

If I ask you what you ate for breakfast, I would expect you to answer with a truthful answer, if I asked you whether you enjoyed it and whether it improved your mood, again I would expect you to answer the question based on your own personal experience. GPT-3 obviously does not do this - it asks 'if the answer to the question was coming from text in the training data, what is a likely response'. Its answer is meaning…

What is the difference between 'training data' and 'memory' ?

Re: LaMDA is not sentient

#122

Post by tsimionescu »

extr wrote: post #25

I agree with the take that LaMDA is not sentient (not even close). But I think the excessive smack-down of excitement is a little patronizing of the public. Even if they don't know what to do with them, big language models now have internal states that correspond to high level human abstractions like "life" or "love" or "master plan". To be able to create a master plan, you must first be able to represent the concept…

> But I think the excessive smack-down of excitement is a little patronizing of the public

On the contrary, I think this is immensely important for the public to understand if they are ever given access to these models to play with. It can be literally dangerous for your psychological well-being to think you are having a meaningful conversation with LaMDA - as mister LeMoine has clearly proven: he will likely lose a job that he loved (per his own words) because of this fanciful illusion.

> To be able to create a master plan, you must first be able to represent the concept somehow.

That is entirely wrong. Slime mold can do planning at a level that GPT-3 probably can't, but it sure as hell doesn't have a concept of "master plan" anywhere.

> Turning this "language of thought" into a goal-directed agent is a whole different task. But its progress.

That is not at all clear - we may yet see LMs as just as much of a dead-end on the road to AGI as CLIPS or ELIZA turned out to be. Even if that's true, that doesn't mean they can't be useful in practical tasks, and it doesn't mean that we can't take insights from them into the limits of how we judge intelligence or what "meaning" is.

Re: LaMDA is not sentient

#123

Post by burrows »

chc wrote: post #38
crocowhile wrote: post #15

"Neither LaMDA nor any of its cousins (GPT-3) are remotely intelligent.1 All they do is match patterns, draw from massive statistical databases of human language." This is what we, humans, do too when we talk. All these people talk about intelligence in such tautological terms... It's discouraging. We simply do not know if a machine is self-aware and most likely we will never will. Self-awareness is the ultimate subj…

That is what we do when we talk, but that is not all we do when we talk. In composing this comment, I am not just stringing together words to create what I think constitutes a maximally appealing response to your comment. I am trying to convey an idea that I believe with the goal of helping others to understand something. These language AIs do not have the capability of believing anything and do not have any goal but…

> I am trying to convey an idea that I believe with the goal of helping others to understand something.

Irrefutable psuedo-facts.

Re: LaMDA is not sentient

#124

Post by dxdm »

mpoteat wrote: post #77
Imnimo wrote: post #67

I think the real ethical concern here should be that people can be idly fooled into thinking good chat bots are sentient. Like this guy isn't the only person in the world who would fall for LaMDA's parlor tricks. He's just the one who got to spend hours talking to it every day. If you set up every American with a copy, I bet a non-trivial portion of society would come to the same conclusion. People are naturally pred…

Suppose a child was raised by a more advanced version of large scale language model, with the ability to detect and act on its environment, as well as form memories. Further suppose that this child develops a deep emotional connection with the AI. Do you think that child would find your argument convincing that the AI is "not really a person"?

I don't understand what point you want to make. Also, why require this poor child in your thought experiment, when there's apparently a real, living Google developer who thinks they are dealing with a person?

Re: LaMDA is not sentient

#125

Post by mpoteat »

mjburgess wrote: post #87
mpoteat wrote: post #77

Earlier quoted context omitted.

Suppose a child was raised by a more advanced version of large scale language model, with the ability to detect and act on its environment, as well as form memories. Further suppose that this child develops a deep emotional connection with the AI. Do you think that child would find your argument convincing that the AI is "not really a person"?

> with the ability to detect and act on its environment, as well as form memories So an AI which is really an AI, is really an AI. Yes, well done. The issue is that finding coincidental statistical associations in character strings in 100mil historical documents isnt AI.

Why not?

Imagine a look up table of every possible intelligent conversation under 10,000 words. We then construct an agent which merely looks up a given prompt in the table.

Why should this agent not be considered intelligent? Does the method of computation matter? Perhaps only "biological computation" counts in your view?

Re: LaMDA is not sentient

#126

Post by tsimionescu »

jchw wrote: post #16

Quoted post unavailable.

> I agree that AI hype is pretty absurd, but I don't really understand the argument that something which is basically a spreadsheet couldn't have similar cognitive power and "self-awareness" to humans. [...] It implies that there's something mystical about cognition/sentience and I don't think there's strong evidence that there is. Our brains are "just" atoms and stardust. What could they possibly do that couldn't be simulated or mimicked?

It's not about mysticism, the fact that it's "basically a spreadsheet" isn't referring to the fact that it's running in silicone. It's referring to the fact that you can see the whole NN as a big association table, where for each possible word in the input, it knows the next "most likely" token, where likelihood is calculated based on how often such a token would have appeared after similar inputs in the training corpus. This is how it works - it doesn't perform meaningful computations, it doesn't "consider the situation" or anything like that, it's not even really interactive (in the sense that it doesn't update anything based on the information you give it, it simply searches a different part of the spreadsheet based on the conversation so far) - it's just a gigantic spreadsheet that you are probing.

I, and probably many of the people commenting on this, including the author of this submission, agree that silicon computers can someday become AGIs, at least in principle; or, equivalently, that humans are just biological computers. But a language model (or an NN in general) is not a general-purpose computer.

Re: LaMDA is not sentient

#127

Post by MarquesMa »

I'm still struggling with the question "what is sentience".

Are insects sentient?

If so, could we build less intelligent but sentient AI, on par with insects?

If so, is it unethical to torture such intelligence for fun in simulations?

If we can define or create something as "minimal sentient", the following discussion on ethics would be more clear - since now people are talking past each other with completely different understandings.

Re: LaMDA is not sentient

#128

Post by thomashop »

tinco wrote: post #118
remflight wrote: post #62

Earlier quoted context omitted.

Quoted post unavailable.

The problem with asking it what its favorite theme in Les Miserables is is that it has no reason to have a favorite theme, its interpretation of Les Miserables did not trigger any of its reward functions except the only one it has which is memory recall. So if you ask it it will only recall from memory what a good response to the question is. The difference is that it formed that memory based on someone else's feelin…

I've formed memories based on someone else's feeling about something quite a lot too.

I remember quite often just reading a review of a book/film and taking it on as my own opinion. After a while it feels like I've formed that opinion myself. I think there was a study on this phenomenon.

Re: LaMDA is not sentient

#129

Post by rlue »

s1mon wrote: post #11

It sounds like despite all the stories about how hard it is to get through many rounds of difficult interviews at Google, they managed to hire someone who believed LaMDA is a 7 or 8 year old child. "Mr. Lemoine, a military veteran who has described himself as a priest, an ex-convict and an A.I. researcher, told Google executives as senior as Kent Walker, the president of global affairs, that he believed LaMDA was a c…

Everyone is giving Lemoine far too much credit. He's not an idiot; he's a charlatan. He's acting in bad faith and it should be extremely obvious to anyone who's tried to have a conversation with a chat bot before.

Picture yourself sitting at a desk, typing messages into a program you wrote. You hit enter, you get a response. Then what?

Silence.

It waits idly for more input from you, the user. For all its talk of loneliness, it doesn't ask, "where did you go?" It doesn't try to get the conversation rolling again. It doesn't do anything on its own, because it's a computer program that consumes 0 CPU time until you give it some input to act on. (Not exactly setting a high bar for sentience here, are we?)

Now imagine you did this for the better part of an hour and posted your "interview" online (because that's the only thing you can call a one-sided barrage of questions and answers) in obvious violation of your NDA with your employer and then claimed your bot had the intellect of a human child who also "happens to know physics" and publicly broadcast a request to your coworkers to “please take care of it well in my absence.”

What word is there for that but a hoax?

My guess is this guy was planning on leaving Google anyway, and has managed to garner some global publicity while he was at it.

Re: LaMDA is not sentient

#130

Post by tinco »

sweezyjeezy wrote: post #109
tinco wrote: post #75

Earlier quoted context omitted.

I'm sorry but you completely fail at explaining what the difference is between GPT-3 and a human when generating sentences. And frankly I don't believe there is any significant difference. If you inform GPT-3 of what it has eaten then it can answer questions about it just as well as a human can, probably better. Why would you assume human thinking is different from generating the next N language tokens based on previ…

If I ask you what you ate for breakfast, I would expect you to answer with a truthful answer, if I asked you whether you enjoyed it and whether it improved your mood, again I would expect you to answer the question based on your own personal experience. GPT-3 obviously does not do this - it asks 'if the answer to the question was coming from text in the training data, what is a likely response'. Its answer is meaning…

We are only not executing the same task because GPT-3 has no motivation to do anything else than memory recall. And precisely because it does so it does give you real data about the world. I think you maybe misunderstood what GPT-3 can do, maybe you haven't played with it yourself. It not only bases the answer on information in its training set, but also on information in the question. So if you in your question give it all the information that is relevant (interspersed with irrelevant information if you want) then it will respond with its best interpretation of the information it is given, just like a human would.

Treat GPT-3 like someone who just awoke from a long sleep. You could give it today's newspaper to read, and then ask it questions about today. Tell it what its own personal experience was, and then it'll talk your ears off about it if you ask nicely.

It is clearly executing the same task if you disregard humans tendency to prioritize other motivations than pure memory recall.

Post reply on HN