I liked the calculator for text analogy. Not sure about now but not for the reason you say, Jurgen.
An LLM is fully reliable as a piece of technology - it never fails to execute the expected calculations to produce the next token. I'd say this is another example of anthropomorphism. "Failure" means something else in the context of a human-like system. (Which an LLM is not)
When we call LLM hallucinations a failure we assume they were built for truthfulness, which they were not and cannot be... My preferred explicit position is that an LLM produce true things by accident, just as confabulations are as much of an accident. The same calculations get performed at the same scale in exactly the same way - the system functions perfectly without fail.
An LLM with no randomness, 0 temperature, is also perfectly predictable in its responses, so even the "reroll" mystique is taken out of the equation.
It is a reliable calculator that does not solve the problems we try to use it for.
I’m going to steal this to explain LLMs to people in the future.
It is indeed the case that to the LLM both truthful responses and hallucinations look identical. It’s just tokens after tokens. And sometimes the answer happens to be right, and sometimes not.
So we describe LLMs or AI in general with metaphors and analogies? If so then that is literally what we do in trying to describe every new thing we encounter. It's the nature of information and how we process it... by metaphor, metaphor, metaphor, all of it. LLMs are just the latest thing to be managed and described by human metaphor, so our human minds can get a grip on it.
I do really agree that there's an open question with no obvious answer: "What are we trying to use this for??" When we apply LLMs to today's day-to-day problems, we get some weird answers (like AI-generated code that then has to be rewritten by humans because of bugs/inaccuracies).
But I also expect profound long-term evolutions of AI use cases. Some big complex tasks will get really simple, really fast, for AI. (Chess anyone?). And most of the 'aha' moments will be when we apply AI to things we haven't yet really considered. Our limit is not the AI itself, but our ability to imagine what it can do.
I think that's true for a lot of things, we make sense of new things by comparing it to what we already understand.
The comparisons between AI and ourselves also makes sense, because that's what we're trying to replicate: an artificial version of human intelligence. But in doing so, the field of AI has also been able to smuggle a bunch of assumptions in for free.
I thought of another one while reading Genesis (https://tinyurl.com/3pwzj63j): AI as a mirror. It reflects our errors, biases, data, and behavior to us. It directly or indirectly reflects our strengths and weaknesses and sometimes perpetuates or amplifies our weaknesses. The question we need to ask is, does it sometimes amplify our strengths?
It's funny, I touched on some of the same themes in my recent "Are We Even Ready For AI Search?" piece, but without focusing directly on the metaphors. But the premise/red thread itself is similar: Don't outsource your own brain and use AI as a tool rather than a replacement for own research, etc.
I like the phrase "crutch of the mind" ... although AI really can be so much more, if we would just stop misusing it. That's really the problem at the core, IMHO.
As for the "metaphors" mentioned in Mitchell's article... anthropomorphism and personification are literary devices distinct from metaphors. Saying things like "gen AI hallucinates" and "AI agents have goals" are not metaphors. These are phrases that anthropomorphize AI. A metaphor would be something like "Gen AI is our new Mt. Everest and we all want to climb it." A "bicycle for the mind" and your "crutch of the mind" are also metaphors. It's ironic that an article about the semantic disagreements about LLMs would mistake one literary device for another. 😎
I agree with you that AI is never one thing; it means as many things the people using it.
At the same time we are shaped by our tools. And if continue on this path — which I have no doubt we will — everything points to the fact that we’re creating a crutch and not a jetpack for the mind.
On the metaphor-stuff, I’m not sure if I agree. If we call the errors of an LLM ‘hallucinations’, are we not leveraging metaphor? To me, anthropomorphization descibes the human tendency, a psychological phenomenon; I wouldn’t call it a literary device.
Right, AI cannot be "one thing" because the term encompasses so many different forms of "AI," such as generative AI systems. Sadly 'tis very true that most of those who utilize chatbots and image generators are either overutilizing or mis-utilizing them, and are increasingly dependent upon what feels like an easy shortcut, which may carry an unexpected higher cost down the road.
Re: metaphors.... if it were as simple as each having our own definitions of literary devices. Anthropomorphism is indeed a literary device, as is metaphor, simile, irony, and so on. Literary devices naturally play on human psychology, that is how they work. That of course doesn't mean you can't use it in a non fiction essay. I do feel it's important not to blur the lines among all of these literary tools, which help us see clearer in fact, lest we too begin to hallucinate...
There's absolutely nothing impressive about AI. It's not intelligent. Its a glorified algorithm. Its being used by dangerous people for horrible things like Israel is using in its genocide. The analogies et al are meant to distract you from the horrors and environmental damage it causes.
Very intriguing article I'm I'm just getting into understanding ai and I'm a retired physicist and chemist but I'm using AI as a a new way of keeping my mind active and on top of what what is considered to be a new realm of science and I think it's very interesting and compelling so that's all I have to say for now but I'm going to continue to explore AR since I have lots of time in my day Tom Woody
I liked the calculator for text analogy. Not sure about now but not for the reason you say, Jurgen.
An LLM is fully reliable as a piece of technology - it never fails to execute the expected calculations to produce the next token. I'd say this is another example of anthropomorphism. "Failure" means something else in the context of a human-like system. (Which an LLM is not)
When we call LLM hallucinations a failure we assume they were built for truthfulness, which they were not and cannot be... My preferred explicit position is that an LLM produce true things by accident, just as confabulations are as much of an accident. The same calculations get performed at the same scale in exactly the same way - the system functions perfectly without fail.
An LLM with no randomness, 0 temperature, is also perfectly predictable in its responses, so even the "reroll" mystique is taken out of the equation.
It is a reliable calculator that does not solve the problems we try to use it for.
This is a great explanation, Ilia.
I’m going to steal this to explain LLMs to people in the future.
It is indeed the case that to the LLM both truthful responses and hallucinations look identical. It’s just tokens after tokens. And sometimes the answer happens to be right, and sometimes not.
So we describe LLMs or AI in general with metaphors and analogies? If so then that is literally what we do in trying to describe every new thing we encounter. It's the nature of information and how we process it... by metaphor, metaphor, metaphor, all of it. LLMs are just the latest thing to be managed and described by human metaphor, so our human minds can get a grip on it.
I do really agree that there's an open question with no obvious answer: "What are we trying to use this for??" When we apply LLMs to today's day-to-day problems, we get some weird answers (like AI-generated code that then has to be rewritten by humans because of bugs/inaccuracies).
But I also expect profound long-term evolutions of AI use cases. Some big complex tasks will get really simple, really fast, for AI. (Chess anyone?). And most of the 'aha' moments will be when we apply AI to things we haven't yet really considered. Our limit is not the AI itself, but our ability to imagine what it can do.
I think that's true for a lot of things, we make sense of new things by comparing it to what we already understand.
The comparisons between AI and ourselves also makes sense, because that's what we're trying to replicate: an artificial version of human intelligence. But in doing so, the field of AI has also been able to smuggle a bunch of assumptions in for free.
I thought of another one while reading Genesis (https://tinyurl.com/3pwzj63j): AI as a mirror. It reflects our errors, biases, data, and behavior to us. It directly or indirectly reflects our strengths and weaknesses and sometimes perpetuates or amplifies our weaknesses. The question we need to ask is, does it sometimes amplify our strengths?
Melanie Mitchell affected me and my pursuit of a Master's degree in genetic algorithms. Her work is recommended.
It's funny, I touched on some of the same themes in my recent "Are We Even Ready For AI Search?" piece, but without focusing directly on the metaphors. But the premise/red thread itself is similar: Don't outsource your own brain and use AI as a tool rather than a replacement for own research, etc.
I’m going to read that. I think some of us are seeing the same things happening and we’re all searching for the right words to describe it.
I like the phrase "crutch of the mind" ... although AI really can be so much more, if we would just stop misusing it. That's really the problem at the core, IMHO.
As for the "metaphors" mentioned in Mitchell's article... anthropomorphism and personification are literary devices distinct from metaphors. Saying things like "gen AI hallucinates" and "AI agents have goals" are not metaphors. These are phrases that anthropomorphize AI. A metaphor would be something like "Gen AI is our new Mt. Everest and we all want to climb it." A "bicycle for the mind" and your "crutch of the mind" are also metaphors. It's ironic that an article about the semantic disagreements about LLMs would mistake one literary device for another. 😎
I agree with you that AI is never one thing; it means as many things the people using it.
At the same time we are shaped by our tools. And if continue on this path — which I have no doubt we will — everything points to the fact that we’re creating a crutch and not a jetpack for the mind.
On the metaphor-stuff, I’m not sure if I agree. If we call the errors of an LLM ‘hallucinations’, are we not leveraging metaphor? To me, anthropomorphization descibes the human tendency, a psychological phenomenon; I wouldn’t call it a literary device.
Right, AI cannot be "one thing" because the term encompasses so many different forms of "AI," such as generative AI systems. Sadly 'tis very true that most of those who utilize chatbots and image generators are either overutilizing or mis-utilizing them, and are increasingly dependent upon what feels like an easy shortcut, which may carry an unexpected higher cost down the road.
Re: metaphors.... if it were as simple as each having our own definitions of literary devices. Anthropomorphism is indeed a literary device, as is metaphor, simile, irony, and so on. Literary devices naturally play on human psychology, that is how they work. That of course doesn't mean you can't use it in a non fiction essay. I do feel it's important not to blur the lines among all of these literary tools, which help us see clearer in fact, lest we too begin to hallucinate...
There's absolutely nothing impressive about AI. It's not intelligent. Its a glorified algorithm. Its being used by dangerous people for horrible things like Israel is using in its genocide. The analogies et al are meant to distract you from the horrors and environmental damage it causes.
Actually that was using polish from Gemini so my apologies for pausing in the middle okay now I'm done
Very intriguing article I'm I'm just getting into understanding ai and I'm a retired physicist and chemist but I'm using AI as a a new way of keeping my mind active and on top of what what is considered to be a new realm of science and I think it's very interesting and compelling so that's all I have to say for now but I'm going to continue to explore AR since I have lots of time in my day Tom Woody
The previous comment by me was edited by Gemini pro 1.5 using profile anyway done