RE: https://mstdn.ca/@drikanis/116107120926277506

I'd like to comment on the common "AI is just a tool" thing: I'm a woodworker by training & that means a lot of machines - but almost every craftsperson knows how to do their job with hand tools, or "lesser" machines.

Similarly, a writer can write without a text editor - just as well, only slower.

If loss of a tool = loss of your skill & knowledge, then that tool isn't an asset, it's a liability. You're signing over your ability to do business to whoever sells & maintains that tool.

#AI

@jwcph Nothing good will come from AI.

@Ponygirl You know, there's a lot of people who would respond to that with a bunch of hemming & hawing about how useful it can/will be for the right applications - but right now I'd say they have the burden of proof & to my knowledge, they're not lifting it.

I'm with you.

@jwcph @Ponygirl
"AI" is not "AI". I hate that "AI" has become the term people use to refer to ChatGPT or Gemini.

You have to distinguish LLMs and other genAI that are being hyped by big tech from the kind of AI that's being used in science and has been used in science for decades.

For example, I use a neural network model to denoise my astrophotography.

"AI" should never have been made available to the general public. This is a thing for science and science alone.

@ninafelwitch @Ponygirl I don't think We™️ have to distinguish. Whatever useful tools scientists & other highly specialised people have which technically fall under this, you guys can & will keep alive regardless of the flak rightfully directed at the hype version & its downfall (hopefully).

- just as long as You™️ remember to distinguish, which sadly isn't always the case; I work at a tech institute (though not an engineer/scientist myself) where they don't & it's causing real problems...

@jwcph @Ponygirl just like nobody has proven that woodworking machines can be useful, right? 😉 (yes, you fell victim of your own analogy)

It's not even remotely hard to prove that LLMs are useful (refactoring, difficult rebases, generating skeletons for projects, etc), anyone who honestly tried them knows that. Their problems are ethical, political, environmental and financial. LLMs (esp. frontier ones) are unsustainable, whether they work or not.

@creepy_owlet @Ponygirl Funny how all that evidence never turns up in public - you would think those examples were worth sharing...

Maybe also tell you boss; about 90% of leaders asked say they can't see any gains thanks to AI.

@jwcph

Similarly, a writer can write without a text editor - just as well, only slower.AI does with text the same thing it does with images, highly detailed approximates that on closer inspection enter uncanny valley. Having to rewrite an AI text is exactly like rewriting a human written final draft - it basically requires going through the first, second and final drafts. AI only saves time if you don't look too closely and/or aren't familiar with what it's talking about.

@Theriac "AI only saves time if you don't look too closely and/or aren't familiar with what it's talking about." 🧑‍🍳💋
@jwcph Funnily enough I wanted to kind of challenge you with saying that it's unlikely a programmer would be able to write a complex program "by hand" (using assembly) but... a sufficiently motivated programmer probably could. It would be an absolutely miserable experience, you'd have to invent a lot from the first principles, but in the end it's all system calls and the documentation is out there.

@art_codesmith
It wouldn't be miserable at all because good programmers relish and enjoy *problem solving*.

It doesn't matter to me what level of tools I have available, so long as they are suited to the task in hand.

I began hand coding assembly language, so I understand why people might think that's awful, but they aren't good engineers, or have lost sight of what good engineering is about because of the corporate and capitalist bullshit that corruptes everything that is good.

@jwcph

@art_codesmith

The point was, that vibe coders can't "code" at all without AI. And pre-AI, people already wrote software. So ... What was your point again?

@jwcph

@fedithom @jwcph My point is that you can apply similar logic to compilers and programming languages. If you’re proficient at making web apps in Python, you maybe *could* make one in assembly, but it would take a lot of time and effort and, as I said, would probably not be a good experience.
@art_codesmith @fedithom @jwcph this is an interesting point and maybe reflective of why llm adoption has been somewhat less controversial among programmers than writers (or woodworkers); the vast majority of programmers already had a near-total dependence on tools, so another level of abstraction is less of a bridge too far

@disconcision @art_codesmith

But can we keep the distinction between using something as a tool and using something as the only means to get any work done? @jwcph

@fedithom @art_codesmith @jwcph agreed that this seems like a meaningful distinction; im saying that for the vast majority of programmers, compilers fall into the category of 'things without which its not possible to get any work done'. writing any machine code at all is a fairly rare skill, and developing non-trivial applications using it is almost non-existent outside of certain specialized sub-domains. this seems to make programming unlike many other arts/crafts, where its the other way around (only certain specific sub-domains basically require specialized tools; many others are doable by hand by most practitioners)
@fedithom @art_codesmith @jwcph (nb i don't really know how relatively true this is for other crafts in general as opposed to programming. i would assume that somewhat adept at digital painting is probably also decent at hand sketching, but also that many/most painters couldn't make their own paints or brushes. so it likely depends on what part of the skill one considers incidental versus essential)

@art_codesmith @jwcph @fedithom
1. I think it is safe to say that competent #software engineers know their tools and an early step in any non-trivial project is to gather tools or write new ones if needed. But we don’t (and cannot) write all of them from scratch because it is too much to keep in our heads AND there are smarter people out there who’ve already done the work. We can do what we do only by leveraging the work of others.

2. A tool created by automatic programming is just as useful as one created by a human. If you trust it to work in your use case then an AI-created tool is no different.

3. The question to be answered is the same for any software tool: Why do I trust it? If you are super-rigorous then you will want to use a formal logic-checking tool to prove the software is correct. That’s really hard and computationally intractable for non-trivial software.

4. ALL software contains residual errors, but our ways of justifying trust in software are incomplete and involve some kind of inductive leap that in the best case leaves you with a quantifiable idea of the risk of failure.

#AI is just software. Do with it what you do with any other software.

@art_codesmith @[email protected] @jwcph

"If you're not mining and refining the materials and building the chips..."

You're conflating different scenarios to the point of absurdity.

@ricardoharvin @jwcph Maybe? I don't know. It was defnitely not my intention.
Maybe I've read too deep into this but, for me, writing in assembly is the best analogy for woodworking with manual tools.
Using a high-level language would be like working with well-developed power tools.
(Using AI... well, the advocates think that it's like working with a super-fancy programmable machine but the motors are busted and the tolerances are between "frick" and "all".)
@jwcph exactly this! I’ve been trying to get that point across for years but you nailed it much better than me!
@jwcph I second this! 💪🏻 In my line of work I (as the IT-dude 😬 ) am implementing AI-tools in our company, because we in that sense then have "control" of what kind of tools the employees are using. But No, it is a false sense of control as we can not control which other AI-tools are in use and by that the whole landscape of AI is our contemporary WILD West, 🫣 🫣
@aj42 @jwcph As a lower level IT dude, I only wrote the user manual for one such internal thing, and felt much psychic pain while doing it. (The IT dude who actually got it running does seem to be on our side though, and I cannot imagine the psychic pain felt to get such things running.)
@ozzelot @jwcph it is "painful" on top f I took the assignment to write the base for our ai-policy 😬 😬
@jwcph I’m also not buying into the common “tools are neutral, it’s how you use them that is good or bad” argument. Tools are built for a purpose, and some purposes are bad. Take the medieval rack, it was built for torture, an intrinsically bad purpose. Sure, you *could* use it for something else, like a weird coffee table. But that’s really beside the point. One who designs and builds a rack, is building something for intrinsically bad ends.
@jwcph A similar concern is the ongoing availability of a tool. Building up your workflows around a tool with sustainability issues or one which is solely controlled by subscriptions to one manufacturer has hurt other crafts time and time again. (e.g. Adobe products)
@jwcph @redmer I learned how to draw with pen & ink. There’s no undo, every stroke on the paper is final. It teaches how to draw with confidence, and dealing with mistakes. This is a skill that translates to drawing digitally. And it’s noticeable when artists only learned to draw digitally and are too dependent on digital tools like undo or specific brushes.
@jwcph also from a baseline If you apply a skill to a tool you receice a specific output. Which is just not true and cannot be true for any current LLM/AI technology the randomness IS the key function.

@jwcph

> "If loss of a tool = loss of your skill & knowledge, then that tool isn't an asset, it's a liability."

So, dump trucks are a liability?

@Downes So, you're an idiot?
@jwcph What kind of response is that?
@Downes @jwcph I don’t understand what you’re saying here.

@Wifiwits @jwcph

Dump trucks are a tool. If we lose dump trucks, then we no longer have the ability & skills required to move large loads of gravel. Therefore, according to the general principle cited ( "If loss of a tool = loss of your skill & knowledge, then that tool isn't an asset, it's a liability") it follows that dump trucks are a liability.

But, of course, dump trucks are not a liability. They make it possible to do what we could not do before. Same with LLMs.

@Downes @Wifiwits Good Lord, are you *trying* to be this dense, or does it come naturally?
@jwcph @Wifiwits Is this your style? Just insult the other person? That's boring.
@Downes @Wifiwits Only people who go out of their way to deserve it.
@Downes @jwcph that analogy doesn’t work at all. These things are not comparable and I’m sure you know this so I can only assume you’re not contributing in good faith here.

@Wifiwits @jwcph I am aware that dump trucks are not the same as LLMs. However, the original statement said "If loss of a tool..." and both dump trucks and LLMs are tools.

The original principle sounds appealing, but the appeal comes from its generality. But it is too broad. It captures too much. So we have to ask, why would this principle apply to LLMs if it doesn't apply to dump trucks?

Try not to respond with insults. It's far more interesting to actually engage with the point being made.

@Downes @Wifiwits Try reading the original posts & also to not be an arrogant dick. That way you're far less likely to be insulted.
@jwcph @Wifiwits I did read the posts. You're a woodworker. You apply the logic of the tools you use to other things. It doesn't work. You know it, and that's why you're getting defensive and lashing out.

@jwcph yes. That's also why ludite were breaking machine. It's not like they were breaking an electric hole maker because they wanted to keep using the hand crank ones. They were breaking tools that required no knowledge of how to do the work to be used. Put a piece of wood on one side, turn a crank, get a door on the other side.

Nothing to learn.

If you learn nothing, no reason to pay you more as time pass, no reason to keep you if you make trouble.

@jwcph I've had the benefit of being fairly isolated from this kind gross over-dependence, and most of the people I've met who use these tools seem to have a realistic grasp on the scope of the problems they're trying to solve.

I'm glad I got to experience some struggle and growth while developing the more difficult skills of my trade before this crutch existed. The temptation NOT to seems to be pretty poisonous.

@jwcph the employee who focuses on making himself indispensable and irreplaceable is one you must terminate.
@keengrasp So... people are tools to you?
@jwcph no, I’m saying it’s very on-brand absurdist hypocrisy for capitalists to seek out the exact same behaviors from their AI tools that they define as immediate grounds for termination when it’s detected in their human servants. Holy shit - Did you think I’ve ever had employees?? I’m a fucking librarian my guy

@keengrasp Well, in my defense, your OP was pretty terse 😝 thank you for clarifying & I agree.

Of course the reason they want from their AI what they don't want from their employees is that they literally own the AI - what they want, and again I'm not in any way being facetious or hyperbolic, is slaves; all the capability & cognition of humans, none of the freedom, up to & including the freedom to exist.

@jwcph That just validates my opinion on LLMs: they are just a tool, and if you can't code without them you shouldn't depend on them in the first place.
In a way, they are a multiplier: they can make a good coder more efficient, but for someone that doesn't know what they're doing they will just result in a lot more bad output. Just like an efficient saw can help a good woodworker, but also result in a lot more wood scraps if used by an unskilled one.

I do agree that the reliance on a handfull of companies is bad though. Since it takes so much resources, it's not like anyone can build a decent LLM, so the competition just isn't there, unlike other tools where there are usually many good options (including more ethical ones...)
@jwcph It's a good point, but I'm totally using AI to do programming things that I have the ability to do as a programmer. It's a tool.
@jwcph @thomasfuchs @drikanis great point – a tool should enhance your skill, not supplant it!

@jwcph

We used to have a bookcase in every software house with tech docs on it and we normalised that being replaced with online docs.

Then Google-fu became important and knowing the right question to form a query around became a good skill for a young dev.

So we got to the point devs developed skills in asking good questions and identifying good answers. The issue is LLM answers don't have peer review like Stack Overflow and get trusted like official docs.

@jwcph I'm also a woodworker and while I can now do a reasonable amount by hand it's not how I was taught and I work with many, particularly carpenters, who can't do any work without power tools. The vast majority can't sharpen a chisel and have never used a hand plane. I've never seen a tenon saw on site and a panel saw is only for crude cuts. Manual skills are there and buildings get built, often well even, but those skills are expressed through machines almost exclusively.
@lostwax Yeah, but those guys still know how all of that works & would be able to teach temselves the hand process - e.g. sharpening a tool - pretty easily if they either had to or wanted to. Read the OP again & remember, this is ostensible professionals *themselves* saying they don't know how to do their job without a specific tool. If you come across a carpenter who says that, you know they're a bad carpenter.
@jwcph yeah they could learn but if a chippy I'm working with has to take the humps out of a stud frame and their electric planer breaks they are going to go and buy a new one, there won't even be a hand plane on site, let alone the skill to use it. You are right that most of those people could learn hand tool work, but most never have, never will and don't want to.
@lostwax None of that was the point, though.
@jwcph I think it is the point honestly. Those carpenters have signed over their ability to do work to mechanisation and power tools, in the same way a programmer might to an LLM. I actually think they've lost something in that and it's been the subject of a good bit of historical handwringing. But within the industry it's now a completely uncontroversial bargain they've made and was probably inevitable as long as we're operating under capitalism.