LLMs Aren't Inevitable(deadsimpletech.com) |
LLMs Aren't Inevitable(deadsimpletech.com) |
The optimists have been more correct in predicting the capabilities and usefulness of LLMs.
The pessimists seem to have a hard time rationalizing what they thought with what’s actually happening.
So what has an "LLM" truly improved in your life outside of simplistic automations/code generation/textual work? It's definitely not improved support. The only support interaction that I've noted to be somewhat better was one (of a few) with Amazon where the bot was able to refund a purchase without talking to a real person. But even then it was a case wherein a human would have probably done it just as quick. Is there a win there? Sure: for Amazon but it didn't "delight" me. And a chatbot is the lowest of low hanging fruit all the "optimists" claim an LLM can do so much better than humans or deterministic code.
How would you say LLMs are impacting the markets? Do you think they're useful to the economy thus far? Or are you looking at the spend of a small fraction of companies that could cause a broader recession when the cyclical funding dries up? If you're going to go to open weight models - well those don't appear out of thin air. Is your focus on the win/loss truly that myopic?
Again, your statement seems absurd to me. I think most people can agree there are useful modalities with LLMs. But it's always at a cost and I, like the author of the article, don't think that LLMs are as universally useful as many "optimists" believe. But to say that the pessimists can't actually rationalize their thoughts, when you provide none of your own, is truly foot in mouth garbage.
I’m happy to engage in discourse on the impact and risks to environment, society, jobs, mental health, etc. There’s plenty of room for thoughtful exploration there.
The one thing I have no patience for is that LLMs produce nothing of value or have no technical worth.
I have to keep reminding myself though, many of those pessimists are people “getting paid to not understand” as the saying goes.
It took 30 seconds with that Davinci model to predict where we are today, where we’ll be next year, and the eventual destruction if not outright extinction of humanity as we know it. Maybe the future is so grim many are simply in the denial stage still.
People like to talk about code as craft and art but the vast majority of the code that was written has always been mediocre at best. LLMs already beat it.
A focused experienced skilled programmer can do better but the market niche for that kind of code is actually small. Most commercial code just has to be good enough to accomplish a business need.
The more pertinent question in my mind is.. why does it still matter? Your customers don’t care about elegant code, they care about working products. There is some relation to good code and good customer experience e.g in terms of performance, but in general your customers won’t feel the impact from poor code. That assumes both pieces of code are functional though, and you do need to babysit LLMs to make sure it actually remains functional.
At this point I think the only reason someone would do better without an LLM is an unwillingness to use it effectively. Even if you’re going to write all code by hand and only use an agent for validation and debugging, that’s a significant boost.
If you really are on the winning/right/good side of this debate, then just be in it, you don't need to sweat it all so much. We have all heard what you or anyone else are going to say on the matter, trust me.
Why care at this point? Why even give the conversation any air by going and responding with the same lines over and over? Is it really just too much to y'all still that some guy has the wrong opinion on the internet?
This could have been fine maybe two years ago, but saying that in the era when LLMs are solving Millenium Prize Problems is incredibly obnoxious. All those opinionated "AI articles" feel like ragebait at this point.
> Now we come to the question of whether or not the difference that LLMs bring to the table is sufficiently overwhelming to make their adoption inevitable, and on this I think the answer is a clear no. As established above, LLMs just can't do top-flight work at all: at the heights of basically any discipline, we're still very much relying on unaugmented human work.
The clear flaw here is that cost isn't taken into account. A new technology needs to be better than the status quo at a given cost/price point.
Examples: mass produced textiles, photography, IKEA. Ostensibly not the highest quality, but the cheapest at a price point.
I can only conclude from this that I am not at all serious in my field. That's okay. I'm too busy enjoying myself.
The rest of the analysis seems to take it for granted that we have hit a wall and LLMs will improve no further. OK. We'll see how that prediction plays out.
I don't know if "most people" is true, yet. I do think that the opprobrium has been silenced, and in corporate environments that has been systematic and top-down. Tokenmaxxing, leaderboards, monitoring: this is all part of making sure developers do not spend time considering the ethics, because the conclusions have been rendered moot.
Nothing that AI will ever be valuable, everything AI does is steal, nothing it creates is new, everything that the big labs publishes is a scam or hype. Oh and "it's just doing X", where you can substitute "X" with some kind of surface level description of how the output layer in an LLM works.
These tenets are already pretty much formed, all it needs is to be formalized by some kind of name or manifesto similar to Q-anon, and we've arrived at a durable state of permanent denial.
Maybe an Orange Catholic Bible.
Empathy and big-picture thinking over myopic specialization will be the new moat.
Shoshanna Zuboff, “The Age of Surveillance Capitalism”
https://news.microsoft.com/source/1997/12/23/dojs-request-fo...
What do LLMs to? You know this: they are sentence completion engines, that through this process of sentence completion can be used to create software than then "does work". This indirection is critical, it is "how this works" and is the true power of LLMs that is STILL not being recognized. The power of LLMs is their communications capacity, their ability to complete sentences, which is so subtle the entire world of software engineers and artificial intelligence everybody has not identified that simple "variable expansion" as the true power of LLMs. LLMs do not do work, they enable the creation of software that then does work.
With a technology so subtle, want to know what is inevitable? A bifurcation of the population into those that believe in magic and those that understand it well enough to make magic nobody considered possible. Why did I say "magic"? Because people are giving up understanding, and that makes what we do magic.
GPT-6.1 (and Opus 5.5, etc.,) is a solid mid-career pro who can produce outputs very much on par with "good human standards" -- and, in some cases, far better than that.
Whatever in-house dev model they're using at OAI/Anthropic is, by all accounts, legitimately superhuman and capable of solving outstanding technical and mathematical problems with three hours of "Pro thinking."
OP's complaint is foolish, it reads like somebody who has never even used these things and has no idea what their capabilities actually are.
“Can produce”. Is doing a lot of heavy lifting there. It can, but then they also regularly write truly bafflingly bad code that no decent experienced dev would ever produce.
I’ll give you a few examples from last week. I had claude whip up a Kafka consumer. After a lot of back and forth I was pretty confident it worked and then I decided to dig into the code. I spotted some real head scratchers.
When starting up the consumer it wanted to pass a specific rebalance strategy, but then it realized that the library it was using didn’t support that strategy, so it needed to use the default.
Instead of just deleting the config option (and maybe leaving a comment), it created a function that removed that key from the map and used it like:
foo: “bar”,
baz: drop_strategy_key(
strategy: “blort”,
bit: “bang”
)
Another frontier model reviewer was supposed to prevent that kind of thing but it didn’t catch it either.
I had 10 public functions that were only called in tests throughout the whole project. Essentially the tests were useless because they weren’t actually testing anything.
Clearly current frontier models are useful. And maybe they produce code that is better than the median dev by many metrics. But I don’t think they are anywhere near consistently producing better quality than a say a 70the percentile dev with 10 years experience.
And even when the quality is better than the median dev, I’m finding that the bugs are not the kind of thing that humans or LLMs are good at spotting.
yep. This is the choice quote.
Even most humans cannot produce work to the highest human standards. The fact is, most work does not need to be done to the highest human standards, it simply needs to be done to the lowest possible standard that someone would pay for, at the cheapest price.
The consumer chooses the cheapest price. Look at sweat shops, look at chinese manufacturing, look at offshoring. All of these produces an inferior product, and yet they are what gets the most profit and overwhelmingly the largest market share of consumer purchases.
yep. This is the choice quote.
Even most humans cannot produce work to the highest human standards. The fact is, most work does not need to be done to the highest human standards, it simply needs to be done to the lowest possible standard that someone would pay for, at the cheapest price.
The first point is the question of whether an LLM can do work not markedly inferior to that of a human doing the same task or not. This one is debatable. An LLM absolutely cannot produce work to the highest human standards, or even a good human standard at the moment, and it's unlikely that it ever will. Most human work, however, is not done at a good standard. At best it's mediocre, and at worst it's downright horrible. The bulk of human knowledge work done in an awful of settings, then, is unfortunately more or less interchangeable with that of an LLM (seriously, so much human work is just awful: I wish I could say otherwise). On this point, then, LLMs probably pass the criterion.
I wish people who felt this way would just go use the tools they hate so much before passing judgement. I think many of them used it once, in a much worse form, many years ago and not since. Judging these things by the standards of 3 months ago is unfair, let alone the standards of 3 years ago.
So it's not really about trying it out, I think. It's something more emotional.
Many people have incredible ability to ignore evidence.
i think those who had backlashes against machines and automation in the past would be similar tbh.
It's because when people feel, even if remotely, their livelihood going down the drain, they do anything and everything to fight it (even if it would be against the interest of overall progress - nobody wants to be the "sacrifice").
They were not just fighting for their livelihoods, though; the fight was at least as much about dignity and fair treatment.
And they were correct to do so: skilled workers who lost their work and ended up in industrial employment did not have better lives. It took a century to a century-and-a-half for industrial quality of life to get back to pre-industrial quality of life.
Loss of dignity is something that the AI industry is not considering as it renders entire swathes of the economy button-pushers for chatbots.
When loss of livelihood appears it will be too late and at too great a scale to fight back in the workplace; the fightback will be somewhere else.
I haven't seen much splitting into two groups of people who are for and against. It's more that individuals can see upsides - does cool stuff, and downsides - something like a rival lifeform that will be smarter than us.
Personally, I'm against AI but even if it could never be smarter than us and if it were at the state that it was last year, I'd still be against it due to the environmental cost.
Good luck saying LLMs are useful but still a net negative in the filter bubble this author is in. No, you also need to think it has zero utility or you're the enemy.
Not to mention how creepy Sam Altman and Dario Amodei are.
The recent efficiency gains in AI have been unparalleled in the history of computing as far as I know. We often point at how a single computer used to occupy an entire room, and over the course of decades of progress, are now wildly surpassed by computers that fit in our pocket. Inference of frontier language models that used to occupy entire data centers earlier this year (Claude Opus 4.6) are now comfortably surpassed in capability by models that can run on a single high end consumer grade gaming PC. I predict data center construction will collapse not just due to pushback from people who hate them, but also lack of demand from people who actually do use AI. People who like AI tend to like running their AI at home, and now they can.
The most important risks in AI right now are the social ones you mentioned. How does the economy evolve? How do we take advantage of this to maximize joy and love? Perhaps most importantly: how do we avoid (incentivize against) falling into a techno-dictatorship?
Superintelligence 2014 https://en.wikipedia.org/wiki/Superintelligence:_Paths,_Dang...
has now brought out another book on the positives:
Deep Utopia 2024 https://ndpr.nd.edu/reviews/deep-utopia-life-and-meaning-in-...
I was just watching an interview with him:
Nick Bostrom Says We Are Clueless About What's Coming https://youtu.be/J9WBZ8BTLLg https://www.nytimes.com/2026/10/01/opinion/interesting-times... - rattles on a bit
Apparently Yudkowsky was influenced by the 2014 book in coming up with his stuff.
And actually the article was quite interesting, though I believe it to be wrong.
I'll include this excerpt from the essay as an attempt to shame folks who feel the urge to continue to claim that the author is critical of "AI" because they hate the tech and never ever have anything good to say about it:
> I will say that Claude Code is a noticeable step change over the Opencode-based GLM models that I tried a few months ago. Those were bad enough to not be usable for anything serious or of real complexity. Claude Code, by contrast, can do some relatively complex things: it was able to perform the bulk of my website migration from Nuxt 3 to Nuxt 4 and to the newest version of Nuxt-Content for instance, which the GLM models had utterly barfed on. It wasn't perfect: I had to do quite a bit of tinkering after the fact to make things work, but unlike the GLM attempt, Claude did actually save me some time. Claude also did pretty well with adding Schema.org markup to my website and performing a number of other stupid-but-necessary SEO-related tasks on the same website: none of these tasks are really groundbreaking, but they're the kind of relatively complex pain-in-the-ass task that you'd hope we could automate: at this point in my practice, there's little of value that I can glean from doing those tasks by hand. If it's possible to make this kind of tool accessible to people at an acceptable, sustainable cost and without burning too many resources, then, I think that'd be a good thing.
> Claude Code also does quite a bit better than Opencode tools did at CI/CD tasks, IaC in OpenTofu, building Dockerfiles and that kind of task. It's still something I'm quite iffy on, but unlike the GLM models I tried earlier, what Claude Code produces basically works, at least initially. This probably makes it a lot more immediately useful than the GLM models were, as this kind of task is a large part of what tech workers actually do day-to-day, and it's the kind of task that a software engineer might have to perform while not having a very good grasp of the underlying technology.
I recommend folks take 15->30 minutes to actually read the essay, rather than either skimming it or having some LLM "summarize" it. There's a lot of subtlety and nuance that the current generation of lossy compressors is incapable of reliably compacting.
[0] <https://deadsimpletech.com/blog/engineering-judgement-claude...>
That needs a citation, and context. Globally? The USA? The Valley? Where is this number from?
https://blog.jetbrains.com/research/2026/08/ai-coding-agent-...
"90% of professional developers were using AI coding agents at work at least weekly in one form or another (local agents or remote cloud agents), with 68% using them daily."
Re. context:
"The Developer Ecosystem and AI Pulse surveys are localized into eight languages: English, Spanish, Chinese, Japanese, Korean, German, French, and Portuguese. We apply quotas on the required number of responses by region to help achieve accurate global representation. ... The Developer Ecosystem Survey has been statistically reweighted to better represent the global developer population by region, employment status, programming language, and familiarity with JetBrains products"
(The survey was run on Jetbrains instagram page, apparently. I mean… that is a bit of selection and narrowing going on there)
Notice how half of them have heard about Jetbrains!? I don't know any developers I would be confident know about Jetbrains (even among AI-using developers)
(Also the graphs don't really support the idea that 90% are using them meaningfully for software development.)
Also, I'd correct your opening sentence like so:
> This is the thing: they probably AI "summarized" it.
I've seen many "summaries" that have come out of those lossy compressors... when tasked with compressing inputs of any appreciable complexity, the output is often too poor to serve as a summary. What you get sure is summary-shaped, though.
> Most of the comments in this thread are, unfortunately defending the thing that will make them less useful and less relevant...
If your claim here is that most of the comments so far are astroturfers or similar, ignore the rest of this line. If it isn't, I'd recommend that you read TFA in its entirety... or go back and read it more carefully if you already did. A sizeable section of the essay talks about what I think you're gesturing at.
Wow, this was ejected off of the front page. It's currently sitting at page 10 two hours after posting. Based on what I've seen over the years, carefully-written criticism about the products of the LLM industry rarely does do well here... for whatever reason.