> i truly have tried to use llms before, just to try it out. nothing i was going to publish as research, but i really couldn't and it left me never wanting to use an llm again
>
> i don't know how more people don't experience this, but staring at an llms output be confidently wrong eventually to get there in the process, or looking at the commands its running and see some really stupid things its doing drives me crazy to the point i will just stop it before it answers, i cannot handle something being that confidently wrong (even though it'll probably still get there in the end)
When people say stuff like this, that they've used LLMs before and hated it, and then list off a bunch of stuff about how they're confidently wrong and all this other stuff, my only question is when? When was the last time that they used an LLM? three months ago, six months ago, one year ago, two years ago?
I'll call myself a late adopter. I tried early models and wasn't impressed, work has made some things available but wasn't forcing us so Ive been keeping my fingers on the pulse for a bit but watching from the sidelines. In October I decided to give it another shot and Opus 5.5 has blown me away, I've written a whole PR for a sparsely documented project and was like write the PR description for me with like one sentence of background, no Claude.md, no context other than the repo and it completely nailed the description. Most of my apps are in the tens of thousands of lines, so nothing huge but still the capabilities of the frontier models is crazy good. They've come a long way from even a year ago.
This is always the first response to criticism of LLMs, and it's starting to feel banal, because it's a counter that you can pull reflexively from your back pocket at any time without engaging with the actual content of the criticism. I'm not saying it's wrong necessarily, but rather that the added information value of "have you tried GPT Cosmica 15" in a discussion on LLMs is approaches zero. Because it's such a generically applicable response.
Well, it's just a rephrasing of "I don't share this experience at all."
And it's an easy explanation for why one person can be getting "confidently wrong" results still, and another person hasn't experienced it in a long time.
I'm one who hasn't experienced it lately. I haven't even thought about it in quite some time. Maybe a domain thing, maybe I'm not doing fancy enough work, maybe that guy is using some shitty model.
Seconding this. It's come to the point where I tried out gemini (argon) and saw it was giving me a wrong answer and was immediately shocked how far behind google obviously still is.
Last week and tonight I tried asking a frontier model a couple of basic questions about a language with 60-70 million speakers with that I'm an expert on. It gave absolutely brutal, confidently wrong answers. It was completely confused and about the basic structure of the language.
Why is every confident claim about a frontier model being brutally wrong about something presented without any details or evidence? It’s so trivial to share even details like what language was used, but even that had to be obscured for some reason.
I think the places the average person starts out with AI do look like this. Most people are not paying a $20 a month Anthropic subscription just to try it out. They're using free chat apps, reading the garbage llm summary Google puts at the top of every search page, dealing with the low quality tools foisted on them in their email app or word processor.
Have all these gotten a lot better in the last year? Sure! But there's still a several-orders-of-magnitude difference in quality between whatever garbage tier of GPT-6 Luna you can use for free on `duck.ai` (just for one example) and Opus 5.5. Or even GLM 5.3 Flash. It's a way bigger difference than the improvement in ~free models has made up for in the last year, even though those improvements have been substantial.
I think a certain sort of person, if they do try to use something better along with a coding harness, as also immediately inclined to read along with the reasoning traces - and understandably gets freaked out by the nonsense it contains. That's a mistake, I think; not just because models behaving "reasonably" doesn't depend on the rationality of their reasoning traces, but also because I think if you transcribed my brain activity into a serial stream of text, I'm not convinced it would be more coherent.
Every time I get pulled into a conversation with someone who is an extreme AI skeptic in this latter half of 2026, it’s eventually revealed that their only experience is one of:
1. The AI box at the top of Google
2. The voice assistance on their phone when they tried it a long time ago. Some question they asked Siri in 2025 and got a bad answer for.
3. ChatGPT free tier.
For a while you’d find someone who got assigned an AI tool at work but their administrator limited them to the cheapest model for cost savings, but most companies are wising up.
I also talked to a developer last month whose company had never let them use anything beyond Sonnet. When I told him I used Opus for everything his eyes went wide because he assumed I was spending thousands of dollars per month. However now that the latest Sonnet is very good, even this category of exposure is decreasing.
I think in 2027 as advanced models become more common we’ll see less of these claims about AI not working well when they tried it. There will be holdout group of people who refuse to try anything new, but the world is moving on.
just starting to read all these; that specifically was both Claude Opus 4.8 and GPT 5.6 Sol and GPT 6 Astra
the main point of the article isn't this and a lot of people seem to be focusing on it though, it's me describing why i personally don't want to use an LLM
and what i'm describing specifically there is the concept of the CoT and the tools it's executing. i see it doing something stupid and my instinct is to stop it, it does usually eventually get there in the end
as i point out right below, llms are super powerful. i'm not arguing that. i'm arguing that we should stop disregarding ethics for marketing and also generally have fun ourselves without llms in personal projects (not work) specifically.
I've been on leave and since my work subscription would go to waste otherwise I've decided to go completely silly and vibecode (truly vibecode) a big project from scratch. Opus 5.5 with xhigh reasoning was repeated firing off hour-long test runs, of tests it had written itself from scratch on code it had written from scratch, until I finally noticed and said "hey, couldn't we run some of these in parallel?" and then the execution time dropped to under 10 minutes. So yes, even recent models do utterly stupid shit pretty frequently (this isn't the only example from this project, but it's the easiest to explain)
Why does that matter? The argument still stands, they are still dumb. We learned some ways to bypass the dumbness and automate some major workflows, and we still see them fail at random things.
It matters because they used to be just dumb, but now they are at the same time smart and dumb.
I can work on ideas with LLMs, even within my own area of expertise. They know quite a lot, and can take an idea I have and run with it. I would never trust them to be correct, but if I give them a new idea and ask them to work out details and raise any issues they find, I can quickly evaluate their findings, and take them into account when developing the idea further.
I do not do this on all ideas, not even most. Most of the time it is still me, a sheet of paper or some code. But still, comparing to how it was a year ago, it is night and day!
A year ago I would just close the window after the first reply, shaking my head at its stupidity. I can still get that experience by using anything but Opus 5.5 or Fable. I tried the $20 ChatGPT tier last week and was immediately thrown back to that old annoyance! (I am sure the best OpenAI models are fine)
But not getting past the forefront of what's known and written by humans, right? (Aside from filling in math proofs by broad searching and brute forcing known lemmas?)
If $1 million on the line general knowledge question I had to choose between an answer from chatgpt Astra and the answer coming from anyone I know currently (or everyone I know answering if even 1 of them gets it right I win) I'm choosing Astra 10/10 times based on my experience. And I suspect I'd likely leave this game with $10 million.
It matters the same way saying “solar panels are super expensive” has been historically true not too long ago, but prices have fallen so rapidly it’s simply no longer a correct statement - even if you will occasionally come across an overpriced offer here and there.
I feel like we're arguing over semantics. What op quoted, referring to what I imagine were things like: confidently running wrong commands, answering something without looking it up, getting stuck on long scripts and not self-correcting all absolutely still happen, so the statement isn't incorrect.
I would also wager that OOP knows LLM limits quite well, and a more interesting discussion would be for why someone perceives LLMs as dumb as opposed to speculating whether they're aware of the frontier or not.
The people I work with outsourcing their research to LLMs are doing wonders. The people I work with outsourcing the line-by-line level of coding to LLMs are hugely productive. It's honestly amazing to see what is possible.
The people I work with outsourcing their thinking to LLMs are struggling, and if I'm not careful they're creating more work for me in review and feedback. The people I work with outsourcing their codebases entirely to LLMs are producing some of the worst software I have ever used.
> It's been decades since beating the best computer at chess was possible.
LLMs can do an impressive job of writing code that meets some specification-but they still can’t beat humans at deciding what that specification should be. That’s especially true when it comes to non-functional requirements, e.g. what constitutes good coding style or good architectural choices.
> The major difference being that winning a chess game doesn't yield a permanent artifact that is useful, generally.
Chess has a single clearly defined “win condition”; coding doesn’t.
This is exactly it. I do not at all feel superfluous in my coding endeavours these days, even when I don’t write any code. If left to its own devices, Opus 5.5 would still fail at the things I want to achieve. It would produce what I would call the toy rocket, when I want the real NASA design. And it requires constant effort in my part to keep it on track to success.
> They clearly aren't writing for others to read their work.
I don't know, the prose and material is fine in my opinion. I do find the uncapitalized style a bit weird, but it's not something that stops me from reading it.
It's a major readability issue for me. I see it periodically and generally assume that person is writing for themselves. Which is fine. The alternative is they're deliberately trying to make things harder for their reader, which I assume usually is not the case.
i'm not gonna let my typing style be criticised when it's been a typing style i've used for previous very popular (even on HN!) articles about vuln research.
it's a choice of typing this way, i don't like typing super formally, just not my thing, i'd rather type casually and most of my articles are sort of braindumps of what i think as i go, but historically i've been complemented very much on this style and i like doing it
The lack of capitals damages readability for others. It's also a writing style strongly associated with arrogance and pretentiousness; but a specific variety of it. Not surprised you've been complimented on some corners of HN.
Every person I've seen on HN who drops capitals is invariably an arrogant prick about it and other things. It's a strong enough calling card that to write that way and be otherwise would be like having a moustache and explaining everywhere "no, I just really like Charlie Chaplain".
but also, throw a weird variation of the game at the computer and it falls apart. Need something novel? llm's wont cut it. Doubly so if what is being done goes against established conventions, even if those are demonstrably wrong, to keep an llm from undermini g everything you first need to show it all the proof and evidence before it can be trusted to help, which for a complex problem can eaisly be 30-50% of co text limits.
LLM's are good at solved problems applied to known or similar to known applications, otherwise they're not going to match humans.
I feel like the AI backlash is creating a rise in "authenticity signaling", cheap public displays of not using AI. Much like virtue signaling, it's lacking in substance and presents as an annoying affectation to people outside of the target in-group.
Good quality well organised and accessible knowledve on a staggering array of topics is at everyones fingertips via wikipedia. Yet, still, I frequently find myself discussing things in depth with friends and collegues that we could easily just lookup and confirm. We dont do this though, because in the process of trying to establish truth using only what is known to us as a group, often creates far more interesting lines of inquery and insight. These are frequently far more potent and useful than anything we can lookup and sometimes are novel enough we cannot find anything like it online. This can only happen because we all applied ourselves and the cognitive varient we embody.
So, the problem with AI is that it is a regression to some mean of minds + corporate bias introduced. It contributes at best one person worth of input variance and is wholly incapable of contributing efficiently or effectively. AI does not contribute effectively and units of insight offered per word read is abysmal for AI text, and so carries enormoys cognitive burden to refine the slop into the simole coherent point actually made, if there even is one, which you often don't know until the effort in refining has been expended. So, we introduce a lot of labor to discussuons by including AI when best case it may offer about as much variance a single person who is not willfully ignorant can offer.
So, I imagine I am not alone in this, and when I talk with humans, what I want is their mental model of the world to contribute, not some corporate mouthpiece i could query directly if that's what I wanted. AI is not exactly inaccessible.
So, i think most are hypervigilant in pointing out AI in their communities for a variety of reasons , but a common theme will be that those who use AI to do their thinking are a single entity in terms of what they actually add to the discussion, and spread over many people it is the age old spam problem, a single mind having disproportionate signal strength in a discussion where all voices should carry equal energy.
I guess it's a way to convey informality and rejection of "the norm" sacrificing legibility.
It's probably a way to express freedom, like the freedom of just refusing to consume unreadable walls of text that disrespect the reader. Freedom goes both ways.
not OP, but it's much simpler for me: when i edited/reordered a sentence, i hated when i had to fix the capitalization. and it's also faster to type without messing with the shift key.
eventually it became a habit and started to leak into more and more of my writing, including public writing. if i wrote something like a blog post, i would still capitalize it, though.
The journal is about a thing this person does privately for fun.
The writing style is clearly not meant for others to read either. They just did stream of consciousness out of their brain and didn't proofread. It's not meant for external consumption.
Capitalization is something you do out of respect for your reader, to make things easier for them. If you aren't intending others to read your writing, then there's no one to disrespect, so why bother?
Let's say the average sentence is 17.5 words, with the average word length being 4.5 characters. Not capitalizing the first word is only a 1.27% character error rate, on average. Not too bad. ;)
something that does meaningfully affect readability is WRITING IN ALL CAPS. you're almost guaranteed to have encountered the visually distinct shapes of lowercased words thousands more times than you've encountered the shapes of their uppercased counterparts
It makes visually scanning the sentences harder, capital letter is bigger than period. I mean, sure, it is readable, but I do prefer regular text, you know, like in a book or newspaper.
more or less, but a lot of people have been doing this for decades. to do it sanely on a phone requires adding lowercased words to your dictionary like i'm, i'd, i'll, and i've. a lot of us also leave off the trailing period
"for me, at least, the process of getting somewhere is more interesting then the end product." -- that's perfectly fine for a hobby, but not for an engineer job.
The people who have this "hobby" attitude of really appreciating the well made steps are the ones that are making deep advances and truly solid things that are used in the long term.
Not at all. We have lots of products (corrugated steel, yoghurt, microwave ovens) that were part of the process of getting somewhere (else) and became the end product. Lots of engineers involved.
as pointed out in the conclusion! that is about specifically the fact even an engineer, who i assume got into this job because they liked doing it as a hobby, should still have some fun with engineering in their personal projects themselves instead of offloading it to an LLM.
i agree that LLM use in work as a software engineer is fully inevitable at this point, because we are all expected to move so fast
I will frequently ask LLM to do research. But I do that because I'm looking to learn from the process LLM implements, as well.
This became a problem recently when I started using Opus 5.5 for a project. It is able to take hundreds of step to (apparently) flawlessly execute a large/complex part of spec. But very recent updates to VScode has really made it hard to observe the process - by obscuring the reasoning blocks and file diffs.
I think there is a lot of interesting discussion around the topic of offloading thinking to AI and the consequences, but blogs like this don’t scratch that itch. This falls more into the category of virtue signaling and pick-a-side division that does well on social media but has little relevance to real life.
The core piece of this blog post is that the author doesn’t use LLMs for fun work because they do the fun work to have fun. Okay. Nothing new there. I some times grow vegetables myself and eat them instead of buying them from the store because I think it’s fun. I don’t feel the need to go around sneering at people who offloaded all of their vegetable growing to someone else. I also don’t like associating with people who get smugly superior and condescending about their personal gardening.
In fact, even though I enjoy gardening and nerding out about it, I don’t like talking to the gardening snobs who have turned it into a part of their personality and who think it makes them superior to the lowly people who don’t garden.
And that’s the exact vibe I get from the articles like this, where the topic drifts into virtue signaling and criticizing the LLM users.
There have been an exhausting supply of these articles for the past few years, and there’s frequently a section like this in each of them:
> i truly have tried to use llms before, just to try it out. nothing i was going to publish as research, but i really couldn't and it left me never wanting to use an llm again
I won’t judge anybody for not using LLMs or for using LLMs for their fun work. Do whatever you want for fun! Don’t let virtue signaling posts like this one shame you into thinking there’s a right way or wrong way to enjoy your fun project time.
That said, I find that to get good discussion about the pros and cons of AI on our thinking you have to look beyond these people who just don’t get LLMs at all. You can’t really understand this topic until you’ve seen both sides.
I wouldn’t find it interesting to read an article on this topic from someone who never learned to program but only uses LLMs to write code because they do not understand the other side. Imagine seeing “I tried to learn how to code once but it didn’t really click for me. It left me never wanting to write code again.” Then they went on to talk about the brilliance of vibecoding with some collected examples of hand-written code causing problems. This article is essentially that, but from the other perspective.
i mean; no, this isn't the conclusion i want someone to takeaway from this article. the part about me not wanting to use LLMs again is purely so i stop getting the question of "why don't you?" and having to explain each time
the takeaway i want is a. i would like people to still have fun and tinker around with things in their personal time instead of making an LLM do it all themselves b. i think we should stop compromising ethics for marketing
> the part about me not wanting to use LLMs again is purely so i stop getting the question of "why don't you?" and having to explain each time
I know why you included it, but it did reveal your level of experience with LLMs. That’s important context for any discussion like this.
> the takeaway i want is a. i would like people to still have fun and tinker around with things in their personal time instead of making an LLM do it all themselves
I know, you wrote that much in the article. That’s what I was responding to: I don’t especially enjoy when other people start taking strong opinions about how other people should and should not spend their personal hobby time. This is why I stay away from most of the online communities for my hobbies these days: The people with strong opinions about how others should spend their personal fun time and the right and wrong ways to enjoy personal time are really not fun.
It would be one thing to write about why you choose to spend your personal time this way. The part where it turned into disdain and got preachy about how other people weren’t doing the same thing was where you lost me.
author is describing the experience of someone with poor prompting skills. when my prompting was poor, I felt the same way. It’s GI-GO. I worked hard to improve my prompting, using a wider range of subject matter, and got a lot better at using the tool. I don’t have that experience ever now, and it was all about becoming a better tool-user.
When as an industry you are collectively doing a trillion in CapEx every year, you are just going to do your best to brainwash the living hell out of everyone to adopt your crap because you really want your money back. Everyone (and their agency and their ability to think even) be damned. The consequences for society? Whatever. There's money you need to make. It's all about the Benjamins, baby. And so, all day long spam, spam, spam. All we do here at Hacker News is talk about these companies. Even worse in other sides of the Internet, there are practical death cults that have formed around all this. It's madness.
> don't know how more people don't experience this, but staring at an llms output be confidently wrong eventually to get there in the process, or looking at the commands its running and see some really stupid things its doing drives me crazy to the point i will just stop it before it answers, i cannot handle something being that confidently wrong (even though it'll probably still get there in the end)
They do? Internet is filled with people complaining about it
My work pays me the same if I kick off a Codex session and go practice guitar. My workflow now is like, prompt LLM, answer some emails or review some code, get a coffee, see what it barfed out.
I've seen people in the office watching live streams while coding. It seems fine to me if the work is getting done, I like some background noise anyways.
Except in that case, your employer eventually catches up with that and forces you to supervise 10x agents, as you're gradually RLHFing your complete replacement.
I don't know who these people are who are supervising 10 agents, but to me there are diminishing returns. There are people on my team who are producing more PRs than me but they're honestly a huge pain in the ass because half of them are nonsense.
> (even though it'll probably still get there in the end)
Burning up tokens/credits needlessly as it does so. That might be a non-issue for people using LLMs through their work but as someone doing AI on a limited budget that's an issue that drives me up a tree!
Two local models I've tried were dead sure that "How Much Is the Fish" was a 1956 song by Dave Brubeck. When asked why they thought so, both just kinda shrugged. At least that was sort of free but seriously — that's the basic knowledge recall scenario, how did it get this garbled? What else gets so horrendously garbled? How can we even know?
i highly recommend asking for terse or concise responses in whatever your chosen LLM calls its system prompt. you can give a brief explanation of your areas of expertise and say that you don't need those things explained to you. there are limits to how much this stuff works, but it's worth a shot
i don't know how more people don't experience this, but staring at an llms output be confidently wrong eventually to get there in the process, or looking at the commands its running and see some really stupid things its doing drives me crazy to the point i will just stop it before it answers, i cannot handle something being that confidently wrong (even though it'll probably still get there in the end)
Let's talk about being "confidently wrong." I don't know if the original author is reading comments here, but if so: how long has it been since you had the experience above?
Have you noticed it happens less and less often as time goes on? And that the models take less time and fewer trials to "get there in the end?"
Because the rate of change is the important part, not the state of the art at the particular moment in time when you formed your opinion.
I'd argue the improvement has come not so much from fixing hallucinations, but more in getting LLMs to successfully run for longer, which lets them more effectively fix their own hallucinations. This is still an improvement, but at the cost of bigger models, with more tokens, running for longer.
I'd be interested to see benchmarks for this sort of thing plotted against cost, power usage, and elapsed time. If these are all sub-linear then that's great, but if some aren't that could be a real problem for LLMs in the long run.
There are too many variables. Autoregressive LLM architectures are not diffusion architectures are not JEPA architectures. We can't say right now how architectures are going to evolve or which one will become prominent even in the near future. There is still so much efficiency left on the table.
Plotting cost against power usage and elapsed time will tell you very little other than what the lay of the land is today. If you are truly concerned about the long term effects, then you need to account for long term evolution.
fully responded to with a comment above but, Claude Opus 4.8, GPT 5.6 Sol and GPT 6 Astra
also, they all get there in the end (as mentioned), but the CoT and command calls and tool calls they do get to me and my instinct is to just stop it right there
the takeaway i want isn't my personal reason on why i don't use ai but rather why people should still not use ai for personal projects, or hobbies. to have fun tinkering with things and coding still, outside of work!
No? LLMs have been outputting more text ever since chain of thought became a thing, and multiple researchers shown thinking traces are outright confabulations, only serving as more memory for latent processing. The tool calls too I believe have gotten more frequent since labs reward tool use regardless of practical value. The net effect is more total intermediate bullshit, even with more results.
OpenAI's harness can solve frontier math problems but the model alone is still a clofabulation machine. Its "intuition" did for a while get better from sheer scale, but we reached the optimum frontier for non-thinking models years ago.
Okay, well, I can't force you to give a shit, I'm saying whether CoT is deceptive is a pretty important question if major AI labs are making monitoring of it a core safety premise https://web.stanford.edu/~cgpotts/blog/cot/#deceptive
I have the same thought on all these "I don't use AI" posts: I would like to not use an LLM too, but I'd soon be out of a job if I didn't. It's an exhausting place to be and I wish I were closer to retirement.
as mentioned in the article; i totally get using llms for work, but for personal projects i think you should spend some time to tinker with it yourself
> eMudhra. you know, the company who had their CA private key as the number Four. this wasn't found by an ai, but just a human using a old ass library made by google called paranoid_crypto to scan through keys for fun out of curiosity
> The goal is to increase the confidence in cryptography use cases inside and outside Google. With Paranoid, any cryptographic artifact can be tested, but its primary motivation is to detect the usage of weak third party hardware or software black boxes.
So google didn't use their own paranoid_crypto lib to check the key before adding it to the Chrome trusted certs.
The tone of this article is "I guess everyone else is stupid now", but his reason for not using LLMs is that he likes doing it the hard way.
Good for him but I don't think that justifies such a snooty position.
I will agree that it kinda sucks not being able to claim credit for your work. I've vibe coded or vibe finished several apps now that are extremely impressive, better than the existing alternatives. But I feel weird saying "I made this" because it feels like I barely did anything.
So I guess everyone uses IDE to code now. I consider anybody not using vim to be a complete moron. Auto complete and syntax highlighting are for the weak minded. /s
You joke but when I first picked up C, I did find it quite helpful to write it in a plain text editor and write my own Makefiles to understand what project setup the IDEs would abstract away - our uni also had pen-and-paper exams where relying on autocomplete would have been difficult :)
(Likewise I can see "you should have read through your generated code and understood the key points yourself" becoming the rule of thumb for keeping maintainers accountable in a project, keeping the pace of changes manageable, and keeping the codebase from becoming a rat's nest _only_ agents can work with)
Even if we're in the same boat, I'm now actually cheering for AI replace my job, purely so people like you will be out on the streets with me and my "I told you so".
> i truly have tried to use llms before, just to try it out. nothing i was going to publish as research, but i really couldn't and it left me never wanting to use an llm again
>
> i don't know how more people don't experience this, but staring at an llms output be confidently wrong eventually to get there in the process, or looking at the commands its running and see some really stupid things its doing drives me crazy to the point i will just stop it before it answers, i cannot handle something being that confidently wrong (even though it'll probably still get there in the end)
When people say stuff like this, that they've used LLMs before and hated it, and then list off a bunch of stuff about how they're confidently wrong and all this other stuff, my only question is when? When was the last time that they used an LLM? three months ago, six months ago, one year ago, two years ago?
I'll call myself a late adopter. I tried early models and wasn't impressed, work has made some things available but wasn't forcing us so Ive been keeping my fingers on the pulse for a bit but watching from the sidelines. In October I decided to give it another shot and Opus 5.5 has blown me away, I've written a whole PR for a sparsely documented project and was like write the PR description for me with like one sentence of background, no Claude.md, no context other than the repo and it completely nailed the description. Most of my apps are in the tens of thousands of lines, so nothing huge but still the capabilities of the frontier models is crazy good. They've come a long way from even a year ago.
This is always the first response to criticism of LLMs, and it's starting to feel banal, because it's a counter that you can pull reflexively from your back pocket at any time without engaging with the actual content of the criticism. I'm not saying it's wrong necessarily, but rather that the added information value of "have you tried GPT Cosmica 15" in a discussion on LLMs is approaches zero. Because it's such a generically applicable response.
Well, it's just a rephrasing of "I don't share this experience at all."
And it's an easy explanation for why one person can be getting "confidently wrong" results still, and another person hasn't experienced it in a long time.
I'm one who hasn't experienced it lately. I haven't even thought about it in quite some time. Maybe a domain thing, maybe I'm not doing fancy enough work, maybe that guy is using some shitty model.
Seconding this. It's come to the point where I tried out gemini (argon) and saw it was giving me a wrong answer and was immediately shocked how far behind google obviously still is.
Last week and tonight I tried asking a frontier model a couple of basic questions about a language with 60-70 million speakers with that I'm an expert on. It gave absolutely brutal, confidently wrong answers. It was completely confused and about the basic structure of the language.
Which model? What language? What questions?
Do you have a share link to the session?
Why is every confident claim about a frontier model being brutally wrong about something presented without any details or evidence? It’s so trivial to share even details like what language was used, but even that had to be obscured for some reason.
They are still confidently wrong often. It’s harder to spot with Opus 5.5 effort medium+ but AI search overview is rife with this
I think the places the average person starts out with AI do look like this. Most people are not paying a $20 a month Anthropic subscription just to try it out. They're using free chat apps, reading the garbage llm summary Google puts at the top of every search page, dealing with the low quality tools foisted on them in their email app or word processor.
Have all these gotten a lot better in the last year? Sure! But there's still a several-orders-of-magnitude difference in quality between whatever garbage tier of GPT-6 Luna you can use for free on `duck.ai` (just for one example) and Opus 5.5. Or even GLM 5.3 Flash. It's a way bigger difference than the improvement in ~free models has made up for in the last year, even though those improvements have been substantial.
I think a certain sort of person, if they do try to use something better along with a coding harness, as also immediately inclined to read along with the reasoning traces - and understandably gets freaked out by the nonsense it contains. That's a mistake, I think; not just because models behaving "reasonably" doesn't depend on the rationality of their reasoning traces, but also because I think if you transcribed my brain activity into a serial stream of text, I'm not convinced it would be more coherent.
> garbage llm summary Google puts at the top of every search page
I assume they don't want to pay for something that would make search results better rather than worse.
Every time I get pulled into a conversation with someone who is an extreme AI skeptic in this latter half of 2026, it’s eventually revealed that their only experience is one of:
1. The AI box at the top of Google
2. The voice assistance on their phone when they tried it a long time ago. Some question they asked Siri in 2025 and got a bad answer for.
3. ChatGPT free tier.
For a while you’d find someone who got assigned an AI tool at work but their administrator limited them to the cheapest model for cost savings, but most companies are wising up.
I also talked to a developer last month whose company had never let them use anything beyond Sonnet. When I told him I used Opus for everything his eyes went wide because he assumed I was spending thousands of dollars per month. However now that the latest Sonnet is very good, even this category of exposure is decreasing.
I think in 2027 as advanced models become more common we’ll see less of these claims about AI not working well when they tried it. There will be holdout group of people who refuse to try anything new, but the world is moving on.
Microsoft copilot is the other one. Copilot has done so much damage to AI's reputation I think AI companies probably despise Microsoft.
just starting to read all these; that specifically was both Claude Opus 4.8 and GPT 5.6 Sol and GPT 6 Astra
the main point of the article isn't this and a lot of people seem to be focusing on it though, it's me describing why i personally don't want to use an LLM
and what i'm describing specifically there is the concept of the CoT and the tools it's executing. i see it doing something stupid and my instinct is to stop it, it does usually eventually get there in the end
as i point out right below, llms are super powerful. i'm not arguing that. i'm arguing that we should stop disregarding ethics for marketing and also generally have fun ourselves without llms in personal projects (not work) specifically.
I've been on leave and since my work subscription would go to waste otherwise I've decided to go completely silly and vibecode (truly vibecode) a big project from scratch. Opus 5.5 with xhigh reasoning was repeated firing off hour-long test runs, of tests it had written itself from scratch on code it had written from scratch, until I finally noticed and said "hey, couldn't we run some of these in parallel?" and then the execution time dropped to under 10 minutes. So yes, even recent models do utterly stupid shit pretty frequently (this isn't the only example from this project, but it's the easiest to explain)
Why does that matter? The argument still stands, they are still dumb. We learned some ways to bypass the dumbness and automate some major workflows, and we still see them fail at random things.
It matters because they used to be just dumb, but now they are at the same time smart and dumb.
I can work on ideas with LLMs, even within my own area of expertise. They know quite a lot, and can take an idea I have and run with it. I would never trust them to be correct, but if I give them a new idea and ask them to work out details and raise any issues they find, I can quickly evaluate their findings, and take them into account when developing the idea further.
I do not do this on all ideas, not even most. Most of the time it is still me, a sheet of paper or some code. But still, comparing to how it was a year ago, it is night and day!
A year ago I would just close the window after the first reply, shaking my head at its stupidity. I can still get that experience by using anything but Opus 5.5 or Fable. I tried the $20 ChatGPT tier last week and was immediately thrown back to that old annoyance! (I am sure the best OpenAI models are fine)
They are improving quickly
But not getting past the forefront of what's known and written by humans, right? (Aside from filling in math proofs by broad searching and brute forcing known lemmas?)
If $1 million on the line general knowledge question I had to choose between an answer from chatgpt Astra and the answer coming from anyone I know currently (or everyone I know answering if even 1 of them gets it right I win) I'm choosing Astra 10/10 times based on my experience. And I suspect I'd likely leave this game with $10 million.
Peak humans? The best we have to offer?
Or my coworkers?
It matters the same way saying “solar panels are super expensive” has been historically true not too long ago, but prices have fallen so rapidly it’s simply no longer a correct statement - even if you will occasionally come across an overpriced offer here and there.
I feel like we're arguing over semantics. What op quoted, referring to what I imagine were things like: confidently running wrong commands, answering something without looking it up, getting stuck on long scripts and not self-correcting all absolutely still happen, so the statement isn't incorrect.
I would also wager that OOP knows LLM limits quite well, and a more interesting discussion would be for why someone perceives LLMs as dumb as opposed to speculating whether they're aware of the frontier or not.
“They are still dumb” “We learned some ways to bypass the dumbness”
"We learned some ways to bypass the dumbness" "They are still dumb"
The people I work with outsourcing their research to LLMs are doing wonders. The people I work with outsourcing the line-by-line level of coding to LLMs are hugely productive. It's honestly amazing to see what is possible.
The people I work with outsourcing their thinking to LLMs are struggling, and if I'm not careful they're creating more work for me in review and feedback. The people I work with outsourcing their codebases entirely to LLMs are producing some of the worst software I have ever used.
Coding is now a lot like playing chess.
It's been decades since beating the best computer at chess was possible. But people still play, because they find it fun to get better at.
The major difference being that winning a chess game doesn't yield a permanent artifact that is useful, generally.
It's fine this person wants to journal about the things they do privately for fun. They clearly aren't writing for others to read their work.
> It's been decades since beating the best computer at chess was possible.
LLMs can do an impressive job of writing code that meets some specification-but they still can’t beat humans at deciding what that specification should be. That’s especially true when it comes to non-functional requirements, e.g. what constitutes good coding style or good architectural choices.
> The major difference being that winning a chess game doesn't yield a permanent artifact that is useful, generally.
Chess has a single clearly defined “win condition”; coding doesn’t.
This is exactly it. I do not at all feel superfluous in my coding endeavours these days, even when I don’t write any code. If left to its own devices, Opus 5.5 would still fail at the things I want to achieve. It would produce what I would call the toy rocket, when I want the real NASA design. And it requires constant effort in my part to keep it on track to success.
> They clearly aren't writing for others to read their work.
I don't know, the prose and material is fine in my opinion. I do find the uncapitalized style a bit weird, but it's not something that stops me from reading it.
It's a major readability issue for me. I see it periodically and generally assume that person is writing for themselves. Which is fine. The alternative is they're deliberately trying to make things harder for their reader, which I assume usually is not the case.
I still write better code than LLMs. It's just slower, and the quality difference is not usually worth the human effort.
i'm not gonna let my typing style be criticised when it's been a typing style i've used for previous very popular (even on HN!) articles about vuln research.
it's a choice of typing this way, i don't like typing super formally, just not my thing, i'd rather type casually and most of my articles are sort of braindumps of what i think as i go, but historically i've been complemented very much on this style and i like doing it
I don't see many compliments of it here.
The lack of capitals damages readability for others. It's also a writing style strongly associated with arrogance and pretentiousness; but a specific variety of it. Not surprised you've been complimented on some corners of HN.
Every person I've seen on HN who drops capitals is invariably an arrogant prick about it and other things. It's a strong enough calling card that to write that way and be otherwise would be like having a moustache and explaining everywhere "no, I just really like Charlie Chaplain".
>They clearly aren't writing for others to read their work.
The author of the blog post and the OP who submitted the article are the same person.
but also, throw a weird variation of the game at the computer and it falls apart. Need something novel? llm's wont cut it. Doubly so if what is being done goes against established conventions, even if those are demonstrably wrong, to keep an llm from undermini g everything you first need to show it all the proof and evidence before it can be trusted to help, which for a complex problem can eaisly be 30-50% of co text limits.
LLM's are good at solved problems applied to known or similar to known applications, otherwise they're not going to match humans.
Some of us capitalize the first letter of every sentence. Wild.
I feel like the AI backlash is creating a rise in "authenticity signaling", cheap public displays of not using AI. Much like virtue signaling, it's lacking in substance and presents as an annoying affectation to people outside of the target in-group.
Good quality well organised and accessible knowledve on a staggering array of topics is at everyones fingertips via wikipedia. Yet, still, I frequently find myself discussing things in depth with friends and collegues that we could easily just lookup and confirm. We dont do this though, because in the process of trying to establish truth using only what is known to us as a group, often creates far more interesting lines of inquery and insight. These are frequently far more potent and useful than anything we can lookup and sometimes are novel enough we cannot find anything like it online. This can only happen because we all applied ourselves and the cognitive varient we embody.
So, the problem with AI is that it is a regression to some mean of minds + corporate bias introduced. It contributes at best one person worth of input variance and is wholly incapable of contributing efficiently or effectively. AI does not contribute effectively and units of insight offered per word read is abysmal for AI text, and so carries enormoys cognitive burden to refine the slop into the simole coherent point actually made, if there even is one, which you often don't know until the effort in refining has been expended. So, we introduce a lot of labor to discussuons by including AI when best case it may offer about as much variance a single person who is not willfully ignorant can offer.
So, I imagine I am not alone in this, and when I talk with humans, what I want is their mental model of the world to contribute, not some corporate mouthpiece i could query directly if that's what I wanted. AI is not exactly inaccessible.
So, i think most are hypervigilant in pointing out AI in their communities for a variety of reasons , but a common theme will be that those who use AI to do their thinking are a single entity in terms of what they actually add to the discussion, and spread over many people it is the age old spam problem, a single mind having disproportionate signal strength in a discussion where all voices should carry equal energy.
I guess it's a way to convey informality and rejection of "the norm" sacrificing legibility.
It's probably a way to express freedom, like the freedom of just refusing to consume unreadable walls of text that disrespect the reader. Freedom goes both ways.
not OP, but it's much simpler for me: when i edited/reordered a sentence, i hated when i had to fix the capitalization. and it's also faster to type without messing with the shift key.
eventually it became a habit and started to leak into more and more of my writing, including public writing. if i wrote something like a blog post, i would still capitalize it, though.
The journal is about a thing this person does privately for fun.
The writing style is clearly not meant for others to read either. They just did stream of consciousness out of their brain and didn't proofread. It's not meant for external consumption.
Capitalization is something you do out of respect for your reader, to make things easier for them. If you aren't intending others to read your writing, then there's no one to disrespect, so why bother?
> It's not meant for external consumption.
The blog author and HN user who posted are the same person. It was meant for external consumption.
Well, it was written without care for those who might read it at any rate.
.and some are putting the dot at the beginning of sentences with all lower case, like in Lojban .it makes much more sense if you think about it
(but i'm half joking here. normally i just don't capitalize, only the names.)
Hmm, that does compensate somewhat.
Let's say the average sentence is 17.5 words, with the average word length being 4.5 characters. Not capitalizing the first word is only a 1.27% character error rate, on average. Not too bad. ;)
it's an intentional display of informality by the writer, i imagine. i do the same a lot of the time.
It's grammatically incorrect and annoying.
Legitimate curiosity, what about capitalization selection annoys you? It feels like a stylistic choice that has no meaningful effect on readability.
something that does meaningfully affect readability is WRITING IN ALL CAPS. you're almost guaranteed to have encountered the visually distinct shapes of lowercased words thousands more times than you've encountered the shapes of their uppercased counterparts
It makes visually scanning the sentences harder, capital letter is bigger than period. I mean, sure, it is readable, but I do prefer regular text, you know, like in a book or newspaper.
Deliberately making sentences harder to parse (by, e.g., bad grammar, bad spelling, lack of capitalization, lack of vowels, leet speak) is annoying.
Non-capitalization may not seem like a huge decrease in readability, but every little bit counts.
prescriptivisT!
(said in the same tone as "splitter!" in Life of Brian)
Believe it or not, some of us enjoy reading human-generated grammatically incorrect stuff more than correctly generated slop.
no it's not :)
at least, i don't think so heh
I guess we’re so used to autocorrect doing that, so that disabling it is a flex now?
more or less, but a lot of people have been doing this for decades. to do it sanely on a phone requires adding lowercased words to your dictionary like i'm, i'd, i'll, and i've. a lot of us also leave off the trailing period
You guys are using autocorrect?
that's model talk! get 'im boys!
What makes you think Erica is an 'im?
Who the fuck is Erica?
You should probably know?
No?
"for me, at least, the process of getting somewhere is more interesting then the end product." -- that's perfectly fine for a hobby, but not for an engineer job.
The people who have this "hobby" attitude of really appreciating the well made steps are the ones that are making deep advances and truly solid things that are used in the long term.
Not at all. We have lots of products (corrugated steel, yoghurt, microwave ovens) that were part of the process of getting somewhere (else) and became the end product. Lots of engineers involved.
as pointed out in the conclusion! that is about specifically the fact even an engineer, who i assume got into this job because they liked doing it as a hobby, should still have some fun with engineering in their personal projects themselves instead of offloading it to an LLM.
i agree that LLM use in work as a software engineer is fully inevitable at this point, because we are all expected to move so fast
Not everyone uses LLM to think.
I will frequently ask LLM to do research. But I do that because I'm looking to learn from the process LLM implements, as well.
This became a problem recently when I started using Opus 5.5 for a project. It is able to take hundreds of step to (apparently) flawlessly execute a large/complex part of spec. But very recent updates to VScode has really made it hard to observe the process - by obscuring the reasoning blocks and file diffs.
I think there is a lot of interesting discussion around the topic of offloading thinking to AI and the consequences, but blogs like this don’t scratch that itch. This falls more into the category of virtue signaling and pick-a-side division that does well on social media but has little relevance to real life.
The core piece of this blog post is that the author doesn’t use LLMs for fun work because they do the fun work to have fun. Okay. Nothing new there. I some times grow vegetables myself and eat them instead of buying them from the store because I think it’s fun. I don’t feel the need to go around sneering at people who offloaded all of their vegetable growing to someone else. I also don’t like associating with people who get smugly superior and condescending about their personal gardening.
In fact, even though I enjoy gardening and nerding out about it, I don’t like talking to the gardening snobs who have turned it into a part of their personality and who think it makes them superior to the lowly people who don’t garden.
And that’s the exact vibe I get from the articles like this, where the topic drifts into virtue signaling and criticizing the LLM users.
There have been an exhausting supply of these articles for the past few years, and there’s frequently a section like this in each of them:
> i truly have tried to use llms before, just to try it out. nothing i was going to publish as research, but i really couldn't and it left me never wanting to use an llm again
I won’t judge anybody for not using LLMs or for using LLMs for their fun work. Do whatever you want for fun! Don’t let virtue signaling posts like this one shame you into thinking there’s a right way or wrong way to enjoy your fun project time.
That said, I find that to get good discussion about the pros and cons of AI on our thinking you have to look beyond these people who just don’t get LLMs at all. You can’t really understand this topic until you’ve seen both sides.
I wouldn’t find it interesting to read an article on this topic from someone who never learned to program but only uses LLMs to write code because they do not understand the other side. Imagine seeing “I tried to learn how to code once but it didn’t really click for me. It left me never wanting to write code again.” Then they went on to talk about the brilliance of vibecoding with some collected examples of hand-written code causing problems. This article is essentially that, but from the other perspective.
i mean; no, this isn't the conclusion i want someone to takeaway from this article. the part about me not wanting to use LLMs again is purely so i stop getting the question of "why don't you?" and having to explain each time
the takeaway i want is a. i would like people to still have fun and tinker around with things in their personal time instead of making an LLM do it all themselves b. i think we should stop compromising ethics for marketing
> the part about me not wanting to use LLMs again is purely so i stop getting the question of "why don't you?" and having to explain each time
I know why you included it, but it did reveal your level of experience with LLMs. That’s important context for any discussion like this.
> the takeaway i want is a. i would like people to still have fun and tinker around with things in their personal time instead of making an LLM do it all themselves
I know, you wrote that much in the article. That’s what I was responding to: I don’t especially enjoy when other people start taking strong opinions about how other people should and should not spend their personal hobby time. This is why I stay away from most of the online communities for my hobbies these days: The people with strong opinions about how others should spend their personal fun time and the right and wrong ways to enjoy personal time are really not fun.
It would be one thing to write about why you choose to spend your personal time this way. The part where it turned into disdain and got preachy about how other people weren’t doing the same thing was where you lost me.
author is describing the experience of someone with poor prompting skills. when my prompting was poor, I felt the same way. It’s GI-GO. I worked hard to improve my prompting, using a wider range of subject matter, and got a lot better at using the tool. I don’t have that experience ever now, and it was all about becoming a better tool-user.
When as an industry you are collectively doing a trillion in CapEx every year, you are just going to do your best to brainwash the living hell out of everyone to adopt your crap because you really want your money back. Everyone (and their agency and their ability to think even) be damned. The consequences for society? Whatever. There's money you need to make. It's all about the Benjamins, baby. And so, all day long spam, spam, spam. All we do here at Hacker News is talk about these companies. Even worse in other sides of the Internet, there are practical death cults that have formed around all this. It's madness.
> don't know how more people don't experience this, but staring at an llms output be confidently wrong eventually to get there in the process, or looking at the commands its running and see some really stupid things its doing drives me crazy to the point i will just stop it before it answers, i cannot handle something being that confidently wrong (even though it'll probably still get there in the end)
They do? Internet is filled with people complaining about it
> i cannot handle something being that confidently wrong
I don’t know how these people have stayed so oblivious to it happening all around them for decades. LLMs didn’t invent it.
My work pays me the same if I kick off a Codex session and go practice guitar. My workflow now is like, prompt LLM, answer some emails or review some code, get a coffee, see what it barfed out.
Must be nice to not have been forced back into the office. Might as well watch paint dry on the walls while I wait for the LLM barf.
I've seen people in the office watching live streams while coding. It seems fine to me if the work is getting done, I like some background noise anyways.
Except in that case, your employer eventually catches up with that and forces you to supervise 10x agents, as you're gradually RLHFing your complete replacement.
Second order effects exist!
I don't know who these people are who are supervising 10 agents, but to me there are diminishing returns. There are people on my team who are producing more PRs than me but they're honestly a huge pain in the ass because half of them are nonsense.
> (even though it'll probably still get there in the end)
Burning up tokens/credits needlessly as it does so. That might be a non-issue for people using LLMs through their work but as someone doing AI on a limited budget that's an issue that drives me up a tree!
Two local models I've tried were dead sure that "How Much Is the Fish" was a 1956 song by Dave Brubeck. When asked why they thought so, both just kinda shrugged. At least that was sort of free but seriously — that's the basic knowledge recall scenario, how did it get this garbled? What else gets so horrendously garbled? How can we even know?
[dead]
i highly recommend asking for terse or concise responses in whatever your chosen LLM calls its system prompt. you can give a brief explanation of your areas of expertise and say that you don't need those things explained to you. there are limits to how much this stuff works, but it's worth a shot
i don't know how more people don't experience this, but staring at an llms output be confidently wrong eventually to get there in the process, or looking at the commands its running and see some really stupid things its doing drives me crazy to the point i will just stop it before it answers, i cannot handle something being that confidently wrong (even though it'll probably still get there in the end)
Let's talk about being "confidently wrong." I don't know if the original author is reading comments here, but if so: how long has it been since you had the experience above?
Have you noticed it happens less and less often as time goes on? And that the models take less time and fewer trials to "get there in the end?"
Because the rate of change is the important part, not the state of the art at the particular moment in time when you formed your opinion.
I'd argue the improvement has come not so much from fixing hallucinations, but more in getting LLMs to successfully run for longer, which lets them more effectively fix their own hallucinations. This is still an improvement, but at the cost of bigger models, with more tokens, running for longer.
I'd be interested to see benchmarks for this sort of thing plotted against cost, power usage, and elapsed time. If these are all sub-linear then that's great, but if some aren't that could be a real problem for LLMs in the long run.
There are too many variables. Autoregressive LLM architectures are not diffusion architectures are not JEPA architectures. We can't say right now how architectures are going to evolve or which one will become prominent even in the near future. There is still so much efficiency left on the table.
Plotting cost against power usage and elapsed time will tell you very little other than what the lay of the land is today. If you are truly concerned about the long term effects, then you need to account for long term evolution.
fully responded to with a comment above but, Claude Opus 4.8, GPT 5.6 Sol and GPT 6 Astra
also, they all get there in the end (as mentioned), but the CoT and command calls and tool calls they do get to me and my instinct is to just stop it right there
the takeaway i want isn't my personal reason on why i don't use ai but rather why people should still not use ai for personal projects, or hobbies. to have fun tinkering with things and coding still, outside of work!
No? LLMs have been outputting more text ever since chain of thought became a thing, and multiple researchers shown thinking traces are outright confabulations, only serving as more memory for latent processing. The tool calls too I believe have gotten more frequent since labs reward tool use regardless of practical value. The net effect is more total intermediate bullshit, even with more results.
OpenAI's harness can solve frontier math problems but the model alone is still a clofabulation machine. Its "intuition" did for a while get better from sheer scale, but we reached the optimum frontier for non-thinking models years ago.
Who cares if the thinking traces are bullshit? They are not the work product.
It matters for monitorability and debugging, viz. OpenAI's misalignment reports which quote heavily from the CoT https://alignment.openai.com/misalignment-reports/
Sounds like somebody else's problem to me. When I want to see the chain of thought, I use a local model. It's seldom enlightening.
Okay, well, I can't force you to give a shit, I'm saying whether CoT is deceptive is a pretty important question if major AI labs are making monitoring of it a core safety premise https://web.stanford.edu/~cgpotts/blog/cot/#deceptive
I have the same thought on all these "I don't use AI" posts: I would like to not use an LLM too, but I'd soon be out of a job if I didn't. It's an exhausting place to be and I wish I were closer to retirement.
as mentioned in the article; i totally get using llms for work, but for personal projects i think you should spend some time to tinker with it yourself
In which ways are you forced to use it though?
> eMudhra. you know, the company who had their CA private key as the number Four. this wasn't found by an ai, but just a human using a old ass library made by google called paranoid_crypto to scan through keys for fun out of curiosity
https://github.com/google/paranoid_crypto
> The goal is to increase the confidence in cryptography use cases inside and outside Google. With Paranoid, any cryptographic artifact can be tested, but its primary motivation is to detect the usage of weak third party hardware or software black boxes.
So google didn't use their own paranoid_crypto lib to check the key before adding it to the Chrome trusted certs.
The bad keys weren't added to the Chrome trusted certs. They were intermediate keys, signed by a key that was trusted.
The tone of this article is "I guess everyone else is stupid now", but his reason for not using LLMs is that he likes doing it the hard way.
Good for him but I don't think that justifies such a snooty position.
I will agree that it kinda sucks not being able to claim credit for your work. I've vibe coded or vibe finished several apps now that are extremely impressive, better than the existing alternatives. But I feel weird saying "I made this" because it feels like I barely did anything.
Author goes by it/she
So I guess everyone uses IDE to code now. I consider anybody not using vim to be a complete moron. Auto complete and syntax highlighting are for the weak minded. /s
You joke but when I first picked up C, I did find it quite helpful to write it in a plain text editor and write my own Makefiles to understand what project setup the IDEs would abstract away - our uni also had pen-and-paper exams where relying on autocomplete would have been difficult :)
(Likewise I can see "you should have read through your generated code and understood the key points yourself" becoming the rule of thumb for keeping maintainers accountable in a project, keeping the pace of changes manageable, and keeping the codebase from becoming a rat's nest _only_ agents can work with)
If you think it's more fun that way, sure. Sometimes (often?) satisfaction is about friction.
And I've seen the /s, but your comment seems to mock OP's point more than just trying to be funny.
Even if we're in the same boat, I'm now actually cheering for AI replace my job, purely so people like you will be out on the streets with me and my "I told you so".