I Don't Want to Interact with Stochastic Parrots | Hacker News Reader

I Don't Want to Interact with Stochastic Parrots

ploum.net

58 pointsfrizlab65 comments

I'm a 60M, recently retired and I wholly agree with this piece. I was classically educated and trained as a structural engineer before turning to software. My experience throughout my career was to observe at every turn the total absence of an engineering ethos which rests on a deep understanding of theory, materials, process and construction. Vibing software will destroy what little rigor there ever was in software development.

It’s the first comment I read in years that doesn’t go “I’m 60 and I’m back to programming again, and having a lot of fun” which make my eyes roll out of their orbits.

Those posts are so many now that I’m inclined to believe they’re all written by AI bots..

Do those people feel important by objecting on those tools? Like they are above others?

I'm getting downvoted while the article itself got flagged.

I still don't understand given the advancement in those models (both closed and open source) one could have such a reductionist view on them. And my only guess is that they derive the feeling of importance through the act of rebellion. Almost like a teenage rebelling against others.

But it doesn't take much to see this tech is here to stay.

Weak egos. Very little of these blind hatreds are derived rationally. They start with a bruised ego first and work backward to make up reasons. Seeing a lot of crypto dualism too, which is hilarious.

Unironically, yes. It takes courage to take a stance and go against the grain at the cost of social ostracism.

What courage?

Hating AI is the norm right now..all those college students, almost everyone in my circle hates AI..the real courage is to speak the truth and try to be objective..

> Using AI is like a shameful drug addiction or a weird sex fetish

This is unfair to drug users and perverts

"Flagged"?

How is this post breaking any guidelines? How does a reasoned essay about LLM use not belong on HN?

Something is rotten in the state of Denmark.

This is how I've felt for a long time now. Sadly, I work at a company whose leadership is incessantly pushing "AI native" and I recently was denied a promotion in part because I don't share our VP's rosy view of AI. "Innovation teams" who generate all of their code with AI get tons of recognition while real engineers that are doing real work behind the scenes are ignored by leadership. I keep hoping this bubble will burst and sanity will return, but I fear that the hold AI has on executives' minds may be a permanent one.

I'm very sorry to say it, but even reading LLM-slop is less tiresome, then reading yet another anti-AI slop article. It's a pretty low bar.

I could find 10 anti-AI posts on HN and they all will contain points from a very narrow pool of anti-AI takes. I mean, how many more times do you need to repeat them?

> stochastic algorithms trained to produce tons of "average," non-human content

As we can see, humans are also quite capable of producing "average" content. And by "average" I don't mean "average quality", I mean "average opinion of a certain group".

> I mean, how many more times do you need to repeat them?

Exactly! It's not as if humans are special (though LLM's certainly are). Humans, they've been the same for thousands of years, same old stories (there's like only 7 archetypal stories they've ever come up with). Same old poems, love this, romance that blah blah blah!

We certainly don't need yet another anti-AI article, same as all the others. Though as it's written entirely by a human who must have thought about every word, certainly poles apart from an LLM generated article one, I'll give him credit for making the effort.

I knew a lot of people have various blind spots, but I had never expected such a massive wave of dualist thinking in practitioners of an engineering/science field.

"...ven reading LLM-slop is less tiresome, then reading..." You meant "than."

Kinda ironic, that a "stochastic parrot" from TFA probably wouldn't make such mistake :)

So why are you commenting on it?

Because every time I hope to see something new and interesting, and every time I'm disappointed.

The stance is defensible but the article reads like it was written in 2023. Still using the term 'stochastic parrot', in 2026, when century-old mathematical problems are being proven by LLMs almost on a weekly basis doesn't make sense.

The solving of the mathematics problems is largely a marketing stunt. It's a great headline though.

It was also deeply uneconomic, and didn't change the world, or materially advance our understanding. Mostly it's demoralised the mathematicians already working to solve those problems. They're not going to say 'That's great, I can now devote my energies to another problem', they're more likely to look at 10 years of poorly paid intellectually rigorous work and go home and kick the dog. For a headline.

Would these mathematicians have said the same thing if another mathematician with a strong competitive drive has beat them to the timing? And how do these mathematicians think of the proof of the four color theorem?

I suppose we don't have the same definition of 'stochastic parrot'.

HN now has AI/LLM-enabled comment processing. Even if you don’t want to interact with the machine it will interact with you. You will eat ze bugs and you will be happy.

What do you mean by this?

It's a reference to a very old meme. Where "eat the bugs" was a shorthand for the lower classes being fed bugs as a cheap source of protein. Has been re-mixed without attribution so many times that I don't know when it started, even Snowpiercer got in on it back in 2013. https://www.youtube.com/watch?v=J_jK9O4tEiA

I think here it's meant to imply that LLMs will be forced on the population at large as a form of social control.

Presumably that HN has an automated LLM-based filter to moderate rude or irrelevant comments, which is a little irrelevant to the article's wish not to see generated posts or comments, and not without its own problems besides: ML is necessary for large-scale moderation, but there are guidelines (like "don't post AI-generated text") which are IMO necessary but also just definitionally hard to enforce in an automated fashion

That your comment and article will be subject to the ranking and processing of a machine authority and there is nothing you can do about it. So you might think you’re being flagged or shadow banned because someone downvoted you or otherwise. But no, it was an automated process that did it.

If it makes the author happier they can ask dang what is the prompt used for automated HN processing.

> Don't post generated text or AI-edited text. HN is for conversation between humans.

This is HN's newest guideline / rule, seems the opposite to me. Does it still happen, yes, is it as bad as 6 months ago, no.

The most human thing is posting something that breaks the rules.

How is that something that defines humans? The machines break the rules as a matter of fact. It's also really a human deciding to have the machine post, they don't do anything of their own volition, so the person behind it is still the one breaking the rules, non sequitur?

I still don't see it as data point we can differentiate human-ness on.

I don’t think they get it and I don’t think they care either way.

> You cannot assume anything about generated code. Maybe it was simply hallucinated and, by chance, it compiled.

This is, ironically, similar to my mom's anti-mental-healthcare argument that the system producing a good outcome is not evidence that the process is correct.

It's everyone's choice what stances they take, and if being utterly opposed to any kind of AI interaction is their's, that's their choice, but just own that choice without resorting to absurdity to attempt to lend it a false sense of credibility.

LLMs don't produce working code through mere hallucinatory chance, they do it by mathematical approximation, which is no more or less valid than pasting together programs from stack overflow answers.

Idgaf if the person who coded a program had intention when writing a line of code I never see because it's a compiled binary blob, I care that the program works. And frankly, pretending otherwise is pretentious BS.

I'm praying that you're not a developer

Worse, I'm someone who's been reviewing code for a long time. And it's not much different now than it was before.

If you believe that compilation proves that the code is good, and that the only thing you should care about a program is if "it works", I don't think you're a great code reviewer.

And in any case, if the code you had experience with (and produced?) before AIs was stack overflow answers pasted together, it's not surprising that you don't see any worsening.

Btw, if your idea of LLMs is of something doing "mathematical approximation", you have some reading to do; and I won't say what I consider pretentious BS.

I think you're putting words in my mouth.

I believe that 2 programs that execute the same are of equal value, no matter the intentions of the person writing the code (or lack thereof). If you read the article, you'd probably know what I was referring to:

> But at least we know that each line, as bad as it can be, was added by a human for a reason.

That's what immediately preceded the bit I quoted, and the entire point of my critique is that this author is treating human intention in coding as intrinsically valuable.

And rather than snark, I'd love if you actually explained why me calling LLMs "mathematical approximations" of their training data is inaccurate? They're literally weighted vector graphs of their training data.

Do you work for IBM? [1]

[1] In IBM there's a religion in software that says you have to count K-LOCs, and a K-LOC is a thousand line of code. How big a project is it? Oh, it's sort of a 10K-LOC project. This is a 20K-LOCer. And this is 5OK-LOCs. And IBM wanted to sort of make it the religion about how we got paid. How much money we made off OS 2, how much they did. How many K-LOCs did you do? And we kept trying to convince them - hey, if we have - a developer's got a good idea and he can get something done in 4K-LOCs instead of 20K-LOCs, should we make less money? Because he's made something smaller and faster, less KLOC. K-LOCs, K-LOCs, that's the methodology. Ugh anyway, that always makes my back just crinkle up at the thought of the whole thing.

No, never worked for Big Blue. ;)

If I get even a whiff of AI smell from an email I receive, the sender gets permablocked. My attention is too valuable for you to insult me by subjecting me to your slop. Take the time for some human interaction. If it's not worth your time to do so, then it most certainly isn't worth mine, either.

If only it were so simple to block all of my coworkers.

We are all stochastic parrots. Currently, millions of years of evolutionary pressure have produced a better algorithm than a few years of optimization pressure but we’ll see how long that remains true!

Evolution selects for fit organisms that have to survive and reproduce in the world, not language models waiting on prompts. The temptation to analogize technology to biology is just top hard to resist I guess. Problem with analogies and metaphors is they are never accurate, they are only similar in some respects.

Yes, thank-you for informing me I was using an analogy.

I think it is worth asking whether anthropomorphization of LLMs does anything to contribute to a better understanding of them that isn't entirely post-hoc ("of course they were going to find CVEs, solve ARC-AGI...") or if it just ends up reinforcing simplified teleological views of neuroscience, evolution or psychology etc.

Sure, there are potential risks to anthropomorphizing language. There's also potential benefits, in terms of allowing us to talk about an emerging alien intelligence.

FWIIW, I don't think I was anthropomorphizing. Stretching an analogy, yes. Optimization pressure is producing something with at least characteristics of intelligence. Obviously in a very different way than evolution produced the human brain, but the similarities are there.

You are not just a stochastic parrot. Your speech is not just "next likely word", but a product of your internal model of the world.

Evidence suggests LLMs have internal models as well.

I would first like to ask, are you religious? Do you believe in souls? If the answer to both is "No", then I'd like to discuss further.

The answer to both is "No".

Ok, then let us discuss what you define stochastic parrot as.

Happy to go with the first paragraph from the Wikipedia entry: https://en.wikipedia.org/wiki/Stochastic_parrot

Suppose a "parrot" factors a product of two 2048 bit primes? Is it still an entity without understanding of the processes considering the search space size?

Does it have an internal model of what a prime is? If so then it's not parroting. If it doesn't then it is. The whole point is that either they have internal models of the concepts, from experience of those things, or they don't, and are purely working from ungrounded symbols which have no connection to the actual things they're referencing.

So, examples of "Look, it did this!" mean nothing to me. What matters is their internal systems.

What do you think "experience is"? The human brain and body stores experience, like anything else as physical data in itself. So why do you think an LLM can't have experiential knowledge if it is trained to or somehow naturally gains it during training?

An LLM, as the name implies, only has access to words. It can only model things about those words and the relationships between them. It never has access to the actual things the words represent.

So it absolutely has experiential knowledge of words but not of the actual things.

(You can absolutely train models on other inputs. I have no objections to those not being stochastic parrots)

Yes but I meant that for instance, how is experience stored in the humans? It is encoded as data in the brain and or rest of body. So I think its possible even a pure textual LLM may have imbibed experiential data somehow during its training.

But if your claim is at least that you are fine with thinking AI's that also include physical training are not stochastic parrots, then thats good enough for me.

Great. And you felt the need to scream this to the entire world? Look, if you don't like Generative AI, then don't use it. This isn't rocket science, and you're not making yourself look hip, cool, edgy, informed, suave, or anything else by yelling about it. Just go do you. Other people may do likewise. Others still might make different choices. And both cases are fine.

And never mind that the whole premise of this article is simply wrong. Look at the things GenAI systems are doing today and it's clear that they are more than just "parrots" stochastic or otherwise. Not even looking at the Navier-Stokes thing, but there are Youtube videos out there of PhD candidates in math who have tried out various models from a "can this help me with research level mathematics?" perspective multiple times, and quite often the answer is at least a qualified "yes". Does it really make sense to say that a stochastic approximation of "very average" content can be helpful in doing research level mathematics? I'm thinking "no".

My own anecdotal experience also suggests that whatever LLM's do, it amounts to more than being a mere parrot. ChatGPT was incredibly helpful to me over the weekend, for example, getting started on some neuromorphic computing stuff I wanted to do. It was amazingly helpful at debugging circuit problems, helping me setup some LXI/SCPI automation of my test equipment, doing literature reviews, yadda, yadda, yadda. Even better, unlike most human interlocutors I might want to interact with, it is fine with my weird "night owl" hours, happily debugging circuits at 2:00am, 3:00am, and on into the early morning hours. It never gets tired, never belittles me, tolerates my repeated questions, etc.

I get that some people are "anti-AI" for their own reasons, but I hope some of those people will at least stop and try to understand why some of us think modern GenAI systems are about the best thing since sliced bread. shrug

The problem (particularly on places like HN) are that if you are openly Anti-AI the people who disagree will flex their ability to downvote you (which contravenes HN's rules [1] [2])

[1] https://news.ycombinator.com/newsguidelines.html

[2] https://news.ycombinator.com/item?id=49605192 (currently at -1 but went as as low as -7)

You go on other sites like Reddit and posit a question. Someone will inevitably reply "So I asked $insert_AI_bot and it said"

It's like second hand smoke. Those of us who don't want to smoke are still subjected to the externalities of the people who want to smoke in public

Outright refusal of the tech based on outdated takes is such a boring, luddite opinion

He is lamenting the tech in a similar manner to lamenting social media. Social media is also 'tech', and yet I don't think we could poor quite enough scorn on it, it deserves more!

I understand the qualities of LLM's but I do fear that we might lose much which once lost may be gone forever. That gives me pause for thought.

You won't be able to choose at some point, and you probably already read plenty of AI generated stuff, it's a survivorship bias.

I have a SKILL that emulates how my supervisor prefers his answers and papers, he used to get triggered by generated stuff, but now it just writes in a style he likes which I got from a bunch of his papers and everything goes smoother. If you are bottlenecking someone they will just do the same to you.

I think the dev holdouts will have to give in on the PR reviews. They catch things and you are delivering less secure software to real people if the luddite desires trump real results for the sake of some false purity test.

Maybe we just shouldn't fill out every optional text field with as much Claude nonsense as it can possibly fill in there. It really feels like they went from "Co-authored-by" propaganda to "You'll know, don't worry. Just watch these invisible watermarks!"

Here's an AI-generated pull request I made, but you wouldn't know because there's no slop:

https://github.com/numtide/devshell/commit/a67c0f87b63bcbdbd...

It got merged by someone who is too busy to respond to personal emails.

"Don't waste my precious attention" is only getting more real.

> have a SKILL that emulates how my supervisor prefers his answers and papers, he used to get triggered by generated stuff, but now it just writes in a style he likes

So your direct superior asked you not to do something and you still worked your way around his preference while feeling cool about of it? Just to save the time of typing?

Clearly the LLM was more than capable to satisfy him, and his conviction was based on pure bias against the technology rather than being grounded in reality.

The only result is that we both have more time. If he didn't like slop, he would catch the new writing as well and I would have to make an effort there, but clearly he is just molded by bias.

Its like people who claim branded stuff is better when it's the same, or that expensive wine/coffee is better when it's about how you serve it. Cognitive bias and stubbornness.

Reply on news.ycombinator.com