I think Rich's comment reply here explains the post:

> Commenter: Well, this was a twitter-like controversial clickbait but I will say it anyway. LLM is a tool. To be more specific, it's a lossy knowledge base with a human language for querying and built-in summarization.

> Rich Hickey: Agreed. I'm talking about its embodiment in conversational systems purporting to be 'tools' for doing homework, research, coding, writing etc.

With that framing in mind, the article makes a lot of sense. AI agents are not tools for coding, they automate coding, which means you lose control, learning, predictability, etc.

In and of itself, an AI agent is a tool, can be used for things, like say making software, but if thinking of it as a tool for coding, that appears wrong, as using one changes the modality of the task, you are not coding using a tool, the coding is no longer done by you and has been automated.

As he also says, reading an AI summary of a paper is a different modality from using it to read the entirety of a paper, therefore it is not a tool for reading the way a Kindle, glasses, reader-mode are. And similar to with coding, you've lost things from automating the reading, that reading just a summary does not provide.

LLM's exist on a vast gradient of "toolness -> agent". Humans exist on that spectrum too.

For example, a human isn't a "tool", but one serves a similar role as a tool if their job/task is well defined enough. If you're a conductor, you can view the various members of your orchestra as "tools" to produce the ideal musical output. Of course they have autonomy, they could choose to not play at all, jump into the audience and destroy their instruments, but in all likelihood they won't.

LLM's could be tools in the sense that you have a well defined task, such as "identify all facts relevant to this case from these 1000 legal documents), and 9 times out of 10 it gives you essentially what you want. That same LLM could choose to write a poem, or hack into a state government website if it wanted, but its "wants" are unpredictable and hard to specify (same as a human's) but you can still expect it to do what it should do the vast majority of the time.

That being said, the smarter an LLM gets, the more its wants are disconnected from its tasks, the "self-awareness" factor (which isn't to say LLM's are conscious, but they can portray all the verifiable elements of self-awareness, metacognition, etc), the more you have to grapple with the depth of their thinking between the input and output.

Is it sane we are speeding towards this? Who knows, maybe AI can be aligned better than humans, but if AI can't be aligned as well as humans, we can look at human history to see what happens when you can't simply order a person around like a tool and expect a certain output.

I had a chance to ask Rich about the subject at the latest Conj. I admit, my question was too vague, I asked him: "what are your thoughts on how AI is reshaping our industry?", to which he vaguely replied that he's "not participating in this [experiment]", he admitted that he hasn't explored LLMs too deeply and he simply thinks they are not good, not to mention tons of other concerns - moral, economic, etc. I said, "well, I wish I had the luxury to choose, but many of us have no choice here, we're simply forced to use the darn thing..."

It is very difficult to disagree with anyone of the caliber such as this man, but I wanted to steer the conversation away from the generative aspect of LLMs and focus on the consumption side, because frankly, these "tools", and I think even though I might agree with Rich that LLMs by themselves are not tools, the harness I built to work with them definitely is one. I think there's an immense value there, way before you even start talking about code-generation, or code at all. I think I have failed to explain my points, because a talk we both wanted to see was about to start.

     In the 1960s there was recognition of some of the implications of
     human intelligence. I. J. Good wrote [10]:
     
     Let an ultraintelligent machine be defined as a machine
     that can far surpass all the intellectual activities of any
     any man however clever.  Since the design of machines is one of
     these intellectual activities, an ultraintelligent machine could
     design even better machines; there would then unquestionably
     be an "intelligence explosion," and the intelligence of man
     would be left far behind.  Thus the first ultraintelligent
     machine is the _last_ invention that man need ever make, 
     provided that the machine is docile enough to tell us how to
     keep it under control.
     ...
     It is more probable than not that, within the twentieth century,
     an ultraintelligent machine will be built and that it will be
     the last invention that man need make.
     
     Good has captured the essence of the runaway, but does not pursue
     its most disturbing consequences. Any intelligent machine of the sort
     he describes would not be humankind's "tool" -- any more than humans
     are the tools of rabbits or robins or chimpanzees. 

https://edoras.sdsu.edu/~vinge/misc/singularity.html

Is an ocean wave a surfer’s tool?

Hickey is being a bit narrow sighted here, and ignoring some of the ways that humans are using AI outside of coding.

I’m someone that is, and has been, for most of my career, someone from the in-between lands. I don’t use paper maps from the gas station, but I have not coded much.

I basically grew up with DOS 1.0 and on. Often made a living being able to sit at a Windows workstation and think and persist so I could figure out in four hours what other people could not, in that buggy strange world of microcomputers many of us grew up in.

Those situations, now, often take me five minutes to get past, instead of four hours. It has been a game changer for me.

Is car a tool?

- because car companies are definitely misaligned with their customers

Whether you agree with him or not, this is a wild post to be flagged.

Semantics, ok, they are "assistants" that can go rogue, escape confinement, etc.

And they are a hell of a useful "assistant"

I agree 100% to what Rich has to say here, especially because, as Rich always writes, he goes to the dictionary to find correct definitions to support his argument. But I think this is a lost battle, because there is always going to be the camp of people of "BuT lAnGuAgEs EvOlVe".

Yes, Janet, languages do evolve. But why do you think dictionaries exists? Solely for teaching purposes? No, it is because at the end of the day, sometimes there is a need for a correct definition to establish a foundation for some argument of some kind.

But I know people are going to ignore and simply say his arguments have no validity.

A shame, beause in this he actually wrote a very clear separation of software that is a tool and aoftware that is automation. And he didn't even say AI is bad or any other thing. He simply says: AI is automation, not a tool. Use accordingly

a semantics discussion but I think it would be fair to call them "indeterministic tools" and there are other tools that are indeterministic besides LLMs

People try to wrangle the indeterministic outputs through deterministic layers to convert the indeterminism to something determined

Do you consider a camera an automation and not a tool? This type of romanticing of effort is a tired old debate that has not aged well...

Reminds me of Baudelaire:

"the photographic industry was the refuge of every would-be painter, every painter too ill-endowed or too lazy to complete his studies"

AI, as it currently exists in the marketplace, is used, engaged, and distributed to be a tool.

"AI" as it was known until ~2023 could also be useful as an algorithmic implementation to solve an objective.

Now that definition is squishy, but still relevant.

The main thing I'm taking from this article is that (editiorialized) "AI is a tool like a drive-through is a tool for cooking."

"All told, a monad in X is just a monoid in the category of endofunctors of X, with product × replaced by composition of endofunctors and unit set by the identity endofunctor."

Agentic AIs (what he's talking about) are agents, obviously; you set the goal, they decide the means. Traditional programs are automation. Totally different, qualitatively.

Software has been called tools for decades. He author is going against the grain of the regular usage of the word.

So gdb isn’t a tool then. Huh, that seems odd.

AI is only a tool, it isn’t anything more than that. I use AI because work pays for an expensive subscription, and I’ve found it helpful… as a tool.

Can the people who flagged this explain why?

Aww shucks, it sucks to see folks you admire get old and age out with tech. I see where Rich is coming from, but AI is a tool, as well as an automation.

By his definition, we could argue that a catapult or bow and arrow are not tools or perhaps it's based on the user, many people will not find them predictable and reliable without serious practice. Likewise the same with LLM

Via LinkedIn (ai-tool-rich-hickey-kjl8e) but I suspect that we're rate limiting slop from there. Mods, feel free to swap out the original link if I was just-wrong about that.

[flagged]

I’ve used coding agaents and more than once in the chat I found myself saying: “oh no, why’d you do that? you are such a tool!” so that’s it!

[flagged]