Clojurians Log v2

Oh how I wish this place was a community of people sharing this view, instead of what it has become 😢

I’ve been meaning to write a Dumb Software Manifesto for a few years now, and Rich beat me to it.

Also, to quip on an https://notes.zachmanson.com/factory-factory-factory/, LLMs are not hammers, they are general-purpose tool-building factory factory factories

"Dumb Software Manifesto", I love it 😄

Damn, on LinkedIn 😭

Rich is so good at distilling essence. I have to say Huahai's rebuttal is also making some interesting counter points here. I've never agreed though with the tool qualification of AI. I mean, we've never called AI a tool before. Is a classification model API endpoint a tool? If I shove a UI over it and loop it, does it become a tool? I agree with Rich, like if you start calling everything a tool, what's the point? Ya an API endpoint is often useful to me for some use-case. Stripe is nice, does it position itself as a tool for payments? I don't know, maybe it does, but I don't particularly think it's useful to see it as a tool in that way. I think the debate over tool in a sense is also unproductive, the qualities implied are much more useful to talk about and we should just discuss those directly. Rich did a good job at bringing qualities up. I think if we talked about those directly we'd avoid a lot of this semantic debate.

Yeah, this thread could probably have benefited from being split into several threads along the way 😅

@didibus also on gists https://gist.github.com/richhickey/a2e19ef3f6f69ca0cca617a8a0da2268 Speaking of, another note from a few months ago https://gist.github.com/richhickey/ea94e3741ff0a4e3af55b9fe6287887f

@john-j-sullivan That eight month old gist is powerful. Wow.

When Rich brings out the dictionary he means serious business!

An example of tautology, saying lots of things without a meaning. first defining tool should be deterministic and by this definition you remove llm from tool category because they are none deterministic, the question is what are your baseline & theory for defining tool & why should people accept it? and why this distinction is important?

I don't think it's an argument. I think it's a warning.

for what?

For the usual turmoil that comes with mass automation.

And probably to think about the inversion of the relationship. You do not use AI, AI uses you.

@sia.mohammady66 I just think we have to be very careful about calling LLMs tools. They are not tools in the sense a hammer is a tool. The “tool” in this case can change its wielder.

Sort of akin to "the medium is the message". Not exactly a new thing, but the scale and speed of the transformation is.

If anything, they are a political tool for the wealthy.

I'm to have to "yes, and..." that statement, given the amount of capable free models available.

but it is definitely the case in the US, like most tech is.

Yes, but it is the same for any other tools in the history. Like tractor or fire or using money or electricity or ... The current LLM usage has lots of change & side effect specially against white collar workforce

I'd argue there is a qualitative difference between LLMs and electricity. You can't talk to electricity - or for that matter, any technology prior to LLMs.

just consider before and after of it, for any thing like using pc you can have the same argument, pc is not a tool, no tool prior to pc have os and being configurable & ...

Perhaps I'm wrong, but I do think a talking machine/software is a different kind of beast as to its implications. The implications are not technical primarily. This technology has social implications and compared to previous transformations that happened over centuries or decades, this will happen within the next few years. It's happening for mathematics as we speak.

I mean it's not really happening to the mathematicians (a bunch of the proofs were shown to be wrong/bogus the day after the announcement, but nobody will remember that). It's mostly marketed as happening to the mathematicians. Which may lead to the same outcome, unemployment of mathematicians, regardless of actual long term effectiveness.

That's before we got onto the actual theft of people prompts/discussion with the LLM to beat them to their own research.

Tangental question. How are people hosting/running their free models?

@andersmurphy Either through openrouter, I think, or if they have a beefy local system with lots of RAM, entirely locally.

I work in the relevant research area, and here they use openrouter a lot

huggingface too

Thanks. Openrouter isn't what I mean't by hosting (especially since they have been bought by Stripe). I'm interested in what people are doing in the rented server/gpu space and/or locally.

I got the same vibe from his Maybe Not talk: you're not supposed to be “under” the logic / type system, the type system should be a tool

I appreciate he lists both predictable and deterministic. People often argue that LLMs are not deterministic when they really mean not predictable. It's like when people mix easy and simple. What Anthropic/OpenAI sell you is a non-deterministic token service. However, LLMs are deterministic in nature. Since determinism is useful, we will likely get deterministic LLMs in practice once challenges (mostly hardware) are solved, making them more tool-like.

why you think they are deterministic? @kloud

Funnily enough, JVM is also not a tool then. And not predicable either. Can't just look at the code and tell how it will behave - gotta properly profile in real-life-like scenarios.

@sia.mohammady66 They are basically just matrix multiplications, those give you token probabilities and with a deterministic function to pick next token you get deterministic result. The challenges are mostly scheduling in GPUs and floating point associativity. Of course coding harnesses bring additional challenges with external tools, but there are techniques from software reproducibility to help with that.

@p-himik performance optimisations like JIT/query planner are transparent, which is the key. The output doesn't change. Only the speed at which you get it.

@kloud in complexity theory we would have emergent behavior, so even simple mechanism in micro scale could shape different phenomena in the macro scale, deterministic in micro is not equal to deterministic in macro

@andersmurphy Depends on how you define "transparent". An LLM is also transparent in the sense that you can even make one yourself - it doesn't stop it from being an LLM. You can even make it deterministic, and you can do that with a third-party model as well. > The output doesn't change What do you mean? JIT is not deterministic.

And something being transparent (whatever that means) and deterministic still doesn't make something predictable. Even if you're a determinist, some things are impossible to predict in principle, due to the sheer amount of computation required. Same here - even if JIT is deterministic, you cannot reliably predict how it will behave.

You're function doesn't return a different output because it was JIT compiled.

I'm not talking about a function output. I'm talking about JIT output, and properties of that output. Performance can be treated as a characteristic of the function, then JIT does change how the function work.

In any case, to me it seems like an exercise in semantics, more than anything. There are more salient bits in there, but they are delivered through this semantic argument. Not sure semantics is a good vehicle for anything but an entry to a part of a talk. We can sit here all day inventing things that don't correspond to what's written in the article that we still colloquially treat as tools, and maybe even things that aren't tools that do satisfy the criteria mentioned there. The good old "what is fish".

fwiw, what I got from this is the wielding part... tool is something you wield and learn to use to do something, whereas LLMs are automation instead of doing something

About this, @andersmurphy: > I mean it's not really happening to the mathematicians (a bunch of the proofs were shown to be wrong/bogus the day after the announcement, but nobody will remember that). It's mostly marketed as happening to the mathematicians. Which may lead to the same outcome, unemployment of mathematicians, regardless of actual long term effectiveness. I found https://www.youtube.com/watch?v=KqiDqJMxg6o particularly enlightening on that specific topic.

> tool is something you wield But that's also semantics. Do you wield a JDK or is it an automation? (Well, its parts anyway). Same with any other compiler - do you wield it, or do you lose any ability to write and interpret ASM/machine code since you only ask a compiler to make it for you? You can do it for most things we consider tools that aren't as simple as a hammer. A jackhammer - is it a tool because I wield it or is it something else because it masks how hard it is to drive a steel rod into the ground, it prevents me from developing my own muscles, it doesn't prove as much feedback as a regular sledge hammer would?

And you can very much use an LLM as a tool in the sense where it's something you wield, control, direct. If you ask it "do the whole project for me", it's hardly that, of course. If you ask it to "refactor this function in this specific way that satisfies these specific constraints", it looks more like a tool to me. Or "convert this JSON into EDN while expanding the set of types given the perceived semantics of the data". Or anything like it.

it's like the oven in the cooking analogy, you don't wield it but you do use it as part of the making process so different than a knife for example

Not sure what new information that brings. "Wield" is an ambiguous overloaded word, and every other pithy word in this whole discussion is just like that. Would I begin to "wield" an oven if I were strong enough to lift and hold it while it's working? What if it's also small enough for me to hold in one hand? What if it doesn't require an external power supply and I simply crank it to provide enough power? If I made my wood-and-metal hammer fall at the press of a button, does that whole hammer+button+whatever stop being a tool? That's all to say that it's still semantics that by itself is, in my opinion, not just useless but is actually detrimental. "A tool or not a tool", that isn't the question. The question is what a particular thing gives you and what it takes away from you. Tools are tangential.

I agree that this semantic discussion is missing the forest for the trees

but for me the analogy of cooking vs ordering food clicked

https://www.linkedin.com/feed/update/urn:li:ugcPost:7515210020456747008/?dashCommentUrn=urn%3Ali%3Afsd_comment%3A%287515265227567943680%2Curn%3Ali%3AugcPost%3A7515210020456747008%29 was my immediate thought when I first read the post: > It appears a car would satisfy the definition of a ‘tool’ if a human driver remains in control of its operations. But what about fully autonomous/self-driving cars? Those also get people from A to B. What about semi-automating features like automatic transmission, adaptive cruise control, and lane-keeping assistance? At what point does a car stop being a tool? […] Are tools and automation necessarily mutually exclusive, or is "toolness" a spectrum? Not that I disagree with what Rich is saying. I'm realizing now, after reading the comments here, that perhaps I thought he was making an argument https://clojurians.slack.com/archives/C0B7Q4XRM1U/p1791785634930089?thread_ts=1791784545.988029&cid=C0B7Q4XRM1U.

but for me the analogy of cooking vs ordering food clicked

@sia.mohammady66 A system can be deterministic but still seem unpredictable for humans, that's why distinguishing the words is important. To give an example: cellular automatons, very simple deterministic rules (your micro scale), but can exhibit seemingly unpredictable yet deterministic behavior like Rule 30 (macro scale).

Using a repl (tool) to write code will make you a more capable coder. Using an llm (automation) to write code will not make you a more capable coder. It might make you more capable of completing projects that involve producing code even if you personally are not gaining skill at coding.

When I'm working at $DAY_JOB the point of what I'm doing is rarely about being a good coder, but rather about accomplishing a goal (often via producing code). Do I need to be a good coder to do so? Maybe. Can I rely on my past experience coding to get by? Also maybe. Is it a problem that I'm no longer gaining much coding expertise from my work projects? Also maybe. The answers to these questions depend on how far I get into unfamiliar territory with my work, and depend on how good the automation is and becomes, and depend on how much banked expertise I can rely on to guide me.

that, and also I'm quite worried about the future where no one knows about the details, how will young colleagues reach deep levels of technical knowledge. are we absolutely 100% certain that it's not needed anymore? I'm not fully convinced

A good metaphor might be a cnc machine. If you have a good cnc machine do you need to know metalworking? If you have a bad one you definitely need to know metalworking. Using a cnc machine will not teach you metalworking. The design of your parts greatly affects how easily you can use a cnc machine to make good parts, and affects much work needs to be done in post to use the parts. Is it a bad thing if you replace your metalworking staff with cnc operators? If your primary concern is the artistry of metalworking, yes this will be a tragedy. If you care mostly about delivering products then maybe the cnc machine is a great boon to you.

The implications about quality are insufficiently supported. Not every manual programmer is a Michelin star chef. It's very possible (though not guaranteed) to get top quality code from an llm. Quality is achievable through both methods. The devil's in the details.

Rich was kind enough to respond to https://www.linkedin.com/feed/update/urn:li:ugcPost:7515210020456747008?commentUrn=urn%3Ali%3Acomment%3A%28ugcPost%3A7515210020456747008%2C7515265227567943680%29&dashCommentUrn=urn%3Ali%3Afsd_comment%3A%287515265227567943680%2Curn%3Ali%3AugcPost%3A7515210020456747008%29.

Whether you should prefer handcrafted products or automated products depends on how good the craftsmen are, how good the automation is, and what aspects have been automated. Analogizing from food service is not going to tell you the answers to these empirical questions in the realm of software engineering so that style of analogy is a waste of time.

Some people find analogies and metaphors to be useful communication tools. Some don't.

Analogies can be useful for mapping out conceptual relationships but it's easy to misapply instincts that were developed in one domain to another domain where the empirical facts are different so the value judgements and instincts should be different too. The state of automation in food service tells us basically nothing about how much quality difference we should expect between automated and handcrafted software.

That's unnecessary (antecedent message deleted)

One of the things I enjoy about the Clojure community is that one can disagree passionately, engage in discussions and dialogue, while focusing on ideas and maintaining a respectful and cordial attitude towards one another as people. I hope we don't lose that as a community.

I can't speak for anyone else, but I will continue to invest in all those things, and our local Clojure community here in Oslo ☺️ With the industry in turmoil and under great uncertainty, it's more important than ever to build a strong community around some shared values and principles.

The cnc metaphor was just to illustrate that the choice between automation and manual craftsmanship is not always straightforward. There are some cases where automation wins, some where manual craftsmanship wins, some where both approaches are complementary. The point of the metaphor is to illustrate the variety of outcomes that could exist, but isn't very useful for determining which outcome is true in particular situations in another domain. I.e. we should not say things like "automation is successful in manufacturing, therefore it will be successful in software engineering" nor should we say "automation produces unpalatable slop in food service, therefore the same will be true in software". Software is software, not food nor manufacturing, so we need to do the hard work if evaluating the actual automations we have access to in the actual software tasks we are challenged with.

I'm not even sure automation is the right word to describe llms/agents. "Delegation" seems more appropriate. It feels similar to giving tasks to a teammate.

I would say it feels more like giving tasks to an offshore consultancy and you don't even get to know the person who will do the tasks.

Story time: I used to make MMO computer games. The studio I worked for had a proprietary client-server game engine, which consisted of several million lines of C/C++ and inline Assembly, written by hundreds of programmers since 1995. That engine shipped four large game titles. Most of the code was written by programmers who had left the company many years ago, and there were huge parts of the codebase that were largely unknown to the programmers working on those four titles, which were all live service games. The oldest title shipped in 2001 (https://www.anarchy-online.com), and is still live today. A lot of that code was atrocious, and nobody dared to touch it. Practically speaking, there wasn't that much difference between having to work on some of that brittle code written by humans vs. sloppy code generated by an LLM. We humans are perfectly capable of producing unmaintainable and bug-ridden code as well.

@p-himik Is it just trendy for everything to be a tool now? I have never heard or thought of JVM as a tool before?

Cue Oprah Winfrey meme: "You get a tool! And you get a tool! And you! Tools for everyone!"

My take from Rich's article is that tool or not, the qualities of something matter, and you should know what they are, what you gain/lose, how they differ between things, etc. Not everything you find useful has the same qualities. If you look at things traditionally called tools, like a hammer, you see a lot of shared qualities between them which for AI agents you don't, there's more qualities in AI agents that are not in traditional tools, and there are more qualities in traditional tools that are not in AI agents, so they overlap minimally in what inherent qualities they provide. And that's something you should understand.

I specifically liked how he started by showing that if you expand the tool definition to anything useful to you, we are all tools working for some person who sees us as just that, a tool 🔥

@didibus I don't know about trends. But an opener with "X is not a tool" invites a discussion on what "tool" is supposed to mean and how it might contradict other common usages of the word. Hence my mention of "what is fish".

E.g. the manpage of javac describes it as a tool. Do we disagree with that notion and ask them to change the manpage? Is this particular point really important enough to justify the turmoil, especially given the "prior art" of fish and its dead end? "I define tool as X and hence anything that's not X is not a tool" - cool, that's that person's vocabulary now. A different person will use a slightly different vocabulary.

I think this is a perfect example of calculative thinking. We are trying to force a new phenomenon into an old category to avoid fear, but that argument avoids the actual issue. I like Heidegger's point of view that @andersmurphy presented (the relationship is inverted now). There is the concept of Poiesis, which is what we are actually losing here. We are not losing just a skill, but also the depth of experience and an authentic relationship to what we create. In the end, it comes down to pure pragmatism versus deep understanding and community.

@p-himik Yes, debating definitions is always a bit pointless. But there's often a hidden really useful discussion underneath about semantics, not about what names we want to give things, but about understanding the things themselves, their properties, qualities, modalities, etc. And in doing so, also understanding their relationships with other things, their uniqueness, their differences and similarities. The challenge is people often begin to fuse the name as an inherent property. You hear two things both referred to as a tool and start to believe therefore they are very similar, and can be clubbed together, that you can approach each the same way, and reason about them interchangeably, what holds true of one thing called a tool must hold true of other things called tools... I think Rich is unpacking that layer. By looking at the actual properties, modalities, invariants, etc. of the things being discussed, a hammer, a pocket, an agentic harness, you realize that what holds true of a hammer really doesn't hold true of an agentic harness, and vice versa. Therefore you have to be careful, if you think of agentic AI as simply another tool.

> what holds true of a hammer really doesn't hold true of an agentic harness It can hold true just fine if you use it that way. This point has been mentioned so many times, it bothers me that it's seemingly kept being ignored. You don't have to interact with AI by prompting it to do your whole job for you. And you can hold a hammer in such a wrong way that it stops being a tool, regardless of what the word means. Granted, AI invites "holding not like a tool" much more readily than a hammer, but still.

@p-himik I'm curious what you mean? How would you "hold" AI agents so they are more like hammers? But I suspect that, as most are having this conversation, they are talking about the way AI agents are generally being held, which is how their providers Anthropic, OpenAI, SpaceX, etc. are showing people how to "hold them". And I assume the same goes for hammers.

I mentioned it earlier in this thread. People have mentioned similar other things so many times that it gets very tiring to repeat the same thing over and over again.

I think we are saying somewhat the same thing surprisingly: what is or not a tool is less important than how you use each thing and what your expectations are of them. That said, I feel instead of getting into that discussion, you're arguing that it can be used as a tool, which keeps the topic stuck on just a naming debate. What do you mean by using it as a tool? And why do you find that use of it is good to reason about as a "tool" ? Because what is often implied when people say AI is just another tool, is that it doesn't change the modality of software development, that it poses no risk to replace the usefulness of human devs, that using it or not is optional and a personal preference, that it's possible to gain considerable mastery over it which would make you dramatically more proficient than others at using it, etc.

Sorry, I don't have enough energy right now to keep addressing every suspicion, generality, implication, and identical question that already has an answer. Please don't let me keep the topic stuck. > what is or not a tool is less important than how you use each thing and what your expectations are of them Yes, of course. This has been said here and in other channels so, so, so many times.

This has been a LONG thread. I will add... any time someone even hints at a metaphor comparing LLMs to anything remotely human I have an extremely sad feeling. How are we at a place where it even makes sense to compare these things to people of any sort? Even randomly chosen people who happen to use Mechanical Turk? There are still leaving breathing people with lived experience on the other side... these things.. are NOT THAT.

@p-himik I'm just lost at what your stance is and to what exactly you have a contradiction on with regards to Rich's article? I'm sorry, I don't keep a stat counter on everything you say to everyone and in other threads 😛. But repeatedly said does not make it true or correct. I'm also not sure what question you refer too? I don't see people asking questions in here?

> but for me the analogy of cooking vs ordering food clicked There's people making recipes, gathering ingredients, cooking and delivering your food when your order it. It's more akin to outsourcing your software development. Agentic software development feels more like fully automated factory design, where the inputs are up to you (can be blurps, ideas, specs, convos) and the output is software. As of today you are still pretty involved in the maintenance, supervision and development of the factory.

I probably won't stick around for the discussion, but I thought this conversation with Casey Muratori and Demetri Spanos had some interesting thoughts in a similar vein: > Will AI Make Me Worse? [Wading Through AI - Episode 5]: In the fifth episode of Wading Through AI, Demetri and Casey discuss the concerns people have about relying on artificial intelligence. > https://youtu.be/4fxo6zuhP1c?si=Yd4BtMlgfqI-VCKN They frame the conversation as service vs tool rather than automation vs tool.

@p-himik Ok, I reread everything you wrote in this thread, and if I understand, you think the article is just arguing semantics over what things should be considered a tool or not? I can see that read, since he approaches the discussion from this semantic debate. But the article does quickly go deeper then that. At least to me, it brings about many interesting things to consider: 1. Held in your hand, continuous instant feedback, tactile 2. Power, where the energy for it comes from, what's the direction of that power, what it allows you to do (leverage) 3. Control, how much control you have in what its doing 4. You do things with it VS it does things for you 5. Predictable behavior, reliable outcomes, bounded mode of operations, simplicity 6. Learning opportunity of using it about the world, physics, dynamics, yourself, etc. 7. Automation lacks many of those properties 8. LLM changes the modality of what you are doing, reading its 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 9. Etc. That's a lot of substance in my opinion, that's way beyond arguing semantics, and many of them I feel are worth discussing and thinking about. In the conclusion he says: > Automation has intrinsic properties and differences from tools that you ignore at your peril. I understand that as, whatever, call it a tool if you want, but please don't ignore the intrinsic properties and differences that LLMs have over other things that might be thought of as tools.

Thought this comment was also interesting: > 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.

the number of "engineers" on the orange website completely missing the point* has me despairing for the current state of reading comprehension. My theory is prolonged exposure to LLMs causes brain damage. * exhibits, yes these are real comments - gdb isn't a tool?! - what about catapults?

You mean like https://publichealthpolicyjournal.com/mit-study-finds-artificial-intelligence-use-reprograms-the-brain-leading-to-cognitive-decline/ ?

Run into this. https://yyhh.org/blog/2026/10/ai-is-a-tool/

I'm having, again, issues with this twitter-like wider reactions in the internet (not here!) and the need to be right and win arguments. I'd rather read all these questions and statements as food for thought (and like someone said on the channel clojurians has maintained remarkable civility regarding discussing hard topics). It is quite easy to choose a viewpoint and start debate but I suspect that many do not start with a positive outcome in mind, a will to learn, thus things quickly deteriorate to defending your chosen view point. This can lead to a net negatice outcome.

Dialogue > Discussion

I agree. However, I do see value in presenting and defending well-formed arguments. When done right, it's a very efficient way to co-discover truth. But when presenting an argument, it should be done with the intent to have it stress tested and seek new information, while being willing to be convinced to change one's position in light of new information. Arguing with a doxastically closed individual is pointless. When we read something, we may choose to charitably interpret the words as "food for thought," even when we don't enjoy how a person presented their thoughts. What annoys me more than anything else is how so many people seemingly don't know how to dispassionately present a valid and sound argument, nor engage with other peoples' arguments on their own merits. Instead, people resort to personal attacks, etc. Eristic is argumentation aimed at winning rather than discovering the truth, which is what most people colloquially get confused with argumentation.

@stephen676 @paul It depends what exposure and the MIT study is really limited. Should we say that using phones makes people stupid just because some people spend too much time on TikTok? Personally, I use LLMs for learning. I have them generate bugs so I can understand the domain by fixing them. I get them to create quizzes, I build personalized learning for my kids, etc. This wasn't really possible before LLMs. https://www.sciencedirect.com/science/article/pii/S2666920X26000962

I think it's a little premature to say thinking will become a hobby. I think existential questions will be thought about more. Philosophical inquiries into the unknown. But I can imagine that thinking in the mode of a means to an end (work) might come to an end. There will be at least some who will reflect on this transition/transformation.

Just read it as a personal story, this person has stopped to think 😁

I can think of many scenarios even outside of software development which make it seem like thinking has been a lost art for decades, long before LLMs became a thing 😅 The domains of religion, politics, war and crime, etc. A great number of people lack the faculties for critical thinking. I hope there are fewer such people in software development.

It's hard for me to resonate with this kind of posts, when for me personally, AI made me think more. I can produce more software, but still the bottle neck is to understand it and make it up to the quality I would perform or smiliar. Since coding got faster and searching for information across project/web, now the thinking part is almost 100% part of the job.

> I can think of many scenarios even outside of software development which make it seem like thinking has been a lost art for decades, long before LLMs became a thing 😅 The domains of religion, politics, war and crime, etc. A great number of people lack the faculties for critical thinking. I hope there are fewer such people in software development. and for these people regulated AI has the potential to educate them, unregulated like Musk's wet dream will make them even dumber feeding them wathever the master tells it to. For a while grok was so bad because they were trying hard to make it go against facts, but I'm sure they have means now to clean a corpus of facts they want gone and replace them with fabricated ones. Overall coherence might still be a problem but endoctrinated people are already quite fine with cognitive dissonance

I think the author has an IQ below an ostrich.

I do appreciate the Clojure community for managing to engage in some of these hard topics and remain mostly amicable and liking each other, even if it can get heated sometimes 😄

@U0BHKM127SQ It is not a dichotomy. You can both view AI as being incredibly beneficial and transformational, while it has serious negative consequences for human existence. We are all still learning what this is about in so many ways.

yeah this channel is way more salty than the other one 😅

I think the "beneficial" part is still up for debate. It's super impressive tech, it's definitely having a transformational impact, at least on software development. But it's not obvious if that transformation is beneficial, to whom, in the short and the long term, etc. In the small I guess you can say it is beneficial, like I benefit from it for certain tasks I am doing. But in the medium/large, and at longer time horizon, I think it's TBD. We're all hoping it ends up with the more optimistic outcome of increased wealth for all, reclaimed time for living and away from work, new discoveries, cures for diseases, etc. Yet we're all worried it ends up with the pessimistic death to humanity, non-zero chance of complete eradication of our race, etc. 😛 or any of it's lesser negative forms like social collapse, new-feudalism, or even just we devs all lose our jobs or end up being paid way way less.

I like to come to this channel to complain or express those worries, and than go back to the other one to live in the moment, unburden by the future, and just enjoy what cool things you can do with this tech right now while there's still something to enjoy.

The other thing is, if beneficial to software development, at what price - the likes of Anthropic are not making a profit, so prices surely must go up (aside from some cost cutting innovation)... or quality down.

has the price of computers gone up or quality down?

I see your point, but it's funny because the price of computers has been going up last 2 years 😛

Aw, looks like jank has two copilot-assisted commits from folks, from before we had our policy against LLM-generated code from external contributors: https://slopscan.ava.pet/repo/https%3A%2F%2Fgithub%2Ecom%2Fjank%2Dlang%2Fjank/main I don't know what a score of 100 (S) means, though. That could be a lot more clear.

Yeah it could be for sure

Yeah not sure how you go about that. There's going to be some percentage of good PRs that are LLM assisted.

Yeah score is confusing this repo is 100(S) which I assumed was 100(slop). But, it has no traces of sausage factory (and none detected) https://slopscan.ava.pet/repo/https%3A%2F%2Fgithub%2Ecom%2Fandersmurphy%2Fhyperlith/master

Yeah so looking at some other orojects the S is for S-tier high welfare handgrown codes.

This little project was almost fully created with chatgpt. I did some manually cleaning, but I would have guessed at least the readme would have been a sign, but still 100(s) 🤷 https://slopscan.ava.pet/repo/https%3A%2F%2Fgithub%2Ecom%2Fjeroenvandijk%2Fchatgpt%2Doidc/main

Yeah I'd expect readme's and commit messages would be the biggest tell.

But it's probably just scanning for claude.md and authorship

So the slopscan is 100% slop?

Heuristics... Last time I checked, it was effectively impossible to tell by looking strictly at the source code. In the worst-case scenario, this kind of "walls of LLM shame" devolves into a witch hunt of sorts. Not a fan without a proper methodology. Heuristics don't cut it.

I mean a grep for the word probe is pretty reliabe.

Yeah, this is bull. My own repo have a 95+ (A) score, which means it's not slop. And unfortunately, it is: https://slopscan.ava.pet/repo/https%3A%2F%2Fgithub%2Ecom%2Fmauricioszabo%2Fpulsar%2Dvscode%2Dcompat/master