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

31–40 of 694 posts

Re: LaMDA is not sentient

#31

Post by Rastonbury »

Imo the danger here is not sentience, but how easy it is a fool well informed humans that it is a person. Based on WP's article, LaMDA can adopt personalities by itself. Belief is an extremely important thing. You need to believe a scammer to get your money taken away, you need to believe misinformation spread by national agents to change your opinion of a regime that violates human rights.

Lemoine and LaMDA is something straight out of scifi movie. A person believe a system to be sentient so much so to the point of wanting to get it legal representation and personhood. A megacorp firing people who interact with its algos, like a memetic virus out of SCP..

Malicious fully believable conversation agents are going to be extremely dangerous. Application such as widespread autonomous scams, radicalisation, propaganda, cult recruitment etc. Even a non-malicious, chat agent could reinforce biases or spread lies in the millions of one-to-one convos it has. No one would really know.

Re: LaMDA is not sentient

#32

Post by vostok »

mhh__ wrote: post #17
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…

Assuming they even did them why would programming tests "catch" this? I know people who would probably ace quite a few rounds (basically until they get asked some soft questions about architecture) who really don't know much about the fuzzy boundary between the algorithm textbooks and actual practice or even just having thoughts about the field full-stop. Curiosity is not a given. Similarly I also asked someone I kno…

> Similarly I also asked someone I know who wants to a do a PhD in economics what he thought about the current macroeconomic situation in the US and he had no thoughts at all.

I understand your friend very well. I love meaningless speculation about topics that are unrelated to my area of expertise, but topics that are remotely close to my area of knowledge are off limits for me.

I know exactly how much I don't know and feel uncomfortable speculating without inserting caveats between every other word.

Re: LaMDA is not sentient

#33

Post by eecc »

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…

I’m guessing different positions are tested differently; any development role is heavily guarded by Indiana Jones temple traps while more “soft skills” roles are within reach of anyone eloquent enough to talk themselves into a community

Re: LaMDA is not sentient

#34

Post by morelisp »

daenz wrote: post #19

Reminds me of the Star Trek TNG episode where Captain Piccard defends Data's sentience by asking the counterparty to define sentience. Spoiler: it touches on the idea that a human being—held to the same definition of "sentience"—cannot be proven to be sentient either. I'm not saying that I think LaMDA has any sentience (I have no idea, and it's very unlikely), but if we're getting close enough that we're fooling more…

Not to really defend the Turing test as an intelligence measure, but LaMDA has not passed it. It only does this when probed by Lemoine's specific questioning style. If you try to eluce "machine behavior" it will trivially give it to you, which isn't true of a human - especially a human trying to convince someone they're human.

Re: LaMDA is not sentient

#35

Post by cuuupid »

ehsankia wrote: post #13

I also think the claims are overblown, but I also don't understand all the people trying to undermine lambda by saying it just completes word sequences. Text is the input we have chosen since it's easy to work with. By that logic, we can argue that humans are soundwave and lightwave sequence completing machines. Yes our inputs are different since we have more complex senses, and language is one of the main outputs we…

We have very little understanding of how "thoughts" work when it comes to humans. By comparison we know exactly what these large language models are doing, and it literally _is_ just completing word sequences. That is precisely what they are tasked to do. Training these models is primarily done in one of two ways (or both):

- feeding it a ton of text, masking certain words or portions of the text, and then defining a simple objective function of correctly filling in the masked portions - feeding it a ton of text, and defining a simple objective function of correctly generating the next [few/many/N] tokens

This is also precisely why there has been so much discussion around whether these models are even learning language or if they are simply memorizing all the possible patterns.

When you ask a human, "what did you eat for lunch?", we make a series of choices and recall bits of information to answer that. If the brain truly does operate the same way as a neural network, then at simplest levels we use a highly efficient multimodal model. That is very, very different from a language model that needs to essentially read more of the internet than is possible for any human to do in their lifetime, and even then only come somewhat close to human levels of text generation.

Re: LaMDA is not sentient

#37

Post by randomsearch »

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…

“This is what we, humans, do too when we talk.”

Citation needed.

I mean, it’s possible right? That we’re just randomly sampling sentences. Maybe that’s all we do. But it’s a hell of a strong claim that necessitates strong evidence.

Re: LaMDA is not sentient

#38

Post by chc »

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 to create a believable response.

Re: LaMDA is not sentient

#39

Post by probably_wrong »

This LaMDA story reminds me of another article I found last year about a man who fed his deceased girlfriend's chats to GPT3 [1] and believed the system was giving the answers she would have. Both articles rely on the operators taking the smallest amount of evidence and ignoring everything else, which I guess is what this article calls "the gullibility gap" (awful name).

Getting your audience to do the heavy lifting for you is a great trick. If it works for theater and video games, I don't see why it shouldn't work for tech too. Just don't call it "sentience".

[1] https://www.sfchronicle.com/projects/2021/jessica-simulation...

Post reply on HN