Anyone else feel like we're cresting the LLM coding hype curve?
Like a recognition that there's value there, but we're passing the frothing-at-the-mouth stage of replacing all software engineers?
Anyone else feel like we're cresting the LLM coding hype curve?
Like a recognition that there's value there, but we're passing the frothing-at-the-mouth stage of replacing all software engineers?
My opinion swings between hype to hate every day. Yesterday all suggestions / edits / answers were hallucinated garbage, and I was ready to remove the copilot plugin altogether. Today I was stuck at a really annoying problem for hours and hours. For shits and giggles I just gave Claude a stacktrace and a description and let it go ham. It produced an amazingly accurate thought train and found my issue, which was not what I was expecting at all.
I still don't see how it's useful for generating features and codebases, but as a rubber ducky it ain't half bad.
Well part of your problem is you are still using copilot. Its fully outdated compared to claude/codex. This tech moves fast.
Well, I gotta use whatever my organization whitelists and provides me a license for. I do use Claude models inside copilot (for Ask/Edit/Agent mode).
Totally understand. My org only allows copilot by default. I convinced my manager to pay for claude, but it was a serious battle to point out how much better it is.
This is a hundred percent true. I felt the same.
What has helped has been to turn off ALL automatic AI, e.g. auto complete, and bind it to a shortcut key to show up on request... And forget it exists.
Until I feel I need it, and then it's throw shit at the wall type moment but we've all been there.
It does save a lot of time as a google on steroid, and wtf-solver. But it's a tool best kept in its box, with a safety lock.
I've been skeptical about LLMs being able to replace humans in their current state (which has gotten marginally better in the last 18 months), but let us not forget that GPT-3.5 (the first truly useful LLM) was only 3 years ago. We aren't even 10 years out from the initial papers about GPTs.
> was only 3 years ago
That's one way of looking at it.
Another way to look at it is GPT3.5 was $600,000,000,000 ago.
Today's AIs are better, but are they $600B better? Does it feel like that investment was sound? And if not, how much slower will future investments be?
Another way to look at $600B of improvement was whether or not they used the $600B to improve it.
This just smells like classic VC churn and burn. You are given it and have to spend it. And most of that money wasn't actually money, it was free infrastructure. Who knows the actual "cost" of the investments, but my uneducated brain (while trying to make a point) would say it is 20% of the stated value of the investments. And maybe GPT-5 + the other features OpenAI has enabled are $100B better.
> And most of that money wasn't actually money, it was free infrastructure.
But everyone who chipped in $$$ is counting against these top line figures, as stock prices are based on $$$ specifically.
> but my uneducated brain (while trying to make a point) would say it is 20% of the stated value of the investments
An 80% drop in valuations as people snap back to reality would be devastating to the market. But that's the implication of your line here.
And yet, we're clearly way into the period of diminishing returns.
I'm sure there's still some improvements that can be made to the current LLMs, but most of those improvements are not in making the models actually better at getting the things they generate right.
If we want more significant improvements in what generative AI can do, we're going to need new breakthroughs in theory or technique, and that's not going to come by simply iterating on the transformers paper or throwing more compute at it. Breakthroughs, almost by definition, aren't predictable, either in when or whether they will come.
Why are you assuming exponential or even linear growth/improvement?
E.g. OpenAI went from "AGI has been achieved internally" to lying with graphs (where they cut off graphs at 50% or 70% to present minor improvements as breakthroughs).
The growth can easily be logarithmic
Well, when MS give OpenAI free use of their servers and OpenAI call it a $10 billion investment, then they use up their tokens and MS calls in $10 billion in revenue, I think so, yes.
I feel like we need a different programming paradigm that's more suited to LLM's strengths; that enables a new kind of application. IE, think of an application that's more analog with higher tolerances of different kinds of user inputs.
A different way to say it. Imagine if programming a computer was more like training a child or a teenager to perform a task that requires a lot of human interaction; and that interaction requires presenting data / making drawings.
Oracle guided program synthesis. The user creates counterfactuals to the program output and the system tries to change its process to correctly process them.
But how is that better?
As a parent, this sounds miserable.
When people talk about the “AI bubble popping” this is what they mean. It is clear that AI will remain useful, but the “singularity is nigh” hype is faltering and the company valuations based on perpetual exponential improvement are just not realistic. Worse, the marginal improvements are coming at ever higher resource requirements with each generation, which puts a soft cap on how good an AI can be and still be economical to run.
What are you basing that on? Haiku 4.5 just came out and beats Sonnet 4 at a third the cost.
GPT-5 and GPT-5-codex are significantly cheaper than the o-series full models from OpenAI, but outperform them.
I won't get into whether the improvements we're seeing are marginal or not, but whether or not that's the case, these examples clearly show you can get improved performance with decreasing resource cost as techniques advance.
> I won't get into whether the improvements we're seeing are marginal or not
But that's exactly the problem!
Right now, AI performs poorly enough that only a small fraction of users is willing to pay money for it, and (despite tech companies constantly shoving it in everyone's face) a large portion of the user base doesn't even want to adopt it for free.
You can't spend hundreds of billions of dollars on marginal improvements in the hope that it'll hopefully eventually become good enough for widespread adoption. Nobody is going to give OpenAI a trillion dollars to grow their user base 50x over the next 15 years. They are going to need to show significant improvements - and soon, or the bubble will pop.
>When people talk about the “AI bubble popping” this is what they mean.
You mean what they have conceded so far to be what they mean. Every new model they start to see that they have to give up a little more.
Maybe, maybe not, it’s hard to tell from articles like this from OSS projects what is generally going on, especially with corporate work. There is no such rhetoric at $job, but also, the massive AI investment seemingly has yet to shift the needle. If it doesn’t they’ll likely fire a bunch of people again and continue.
It's been less than a year and agents have gone from patently useless to very useful if used well.
Useful if used well as a thought has gone from meaning a replace all developers machine to a fresh out of college junior with perfect memory bot to a will save a little typing if you type out all of your thoughts and baby sit it text box.
I get value from it everyday like a lawyer gets value from LexisNexis. I look forward to the vibe coded slop era like a real lawyer looks forward to a defendant with no actual legal training that obviously did it using LexisNexis.
The trajectory is a replace all developers trajectory, you're just in the middle of the curve wondering why you're not at the end of it.
The funny thing is you're clearly within the hyperbolic pattern that I've described. It could plateau, but denying that you're there is incorrect.
> you're just in the middle of the curve wondering why you're not at the end of it.
You assume the curve is exponential.
We assume the curve is logarithmic.
We are not the same
Where are you employed?
Why you ask a stranger on the internet for PII?
I'm genuinely curious as to what's going through your mind and if people readily give you this.
I suspect you're asking dishonestly but I can't simply assume that.
Every single one of your posts from the past two weeks is hyping up AI or down voted for being highly uninformed about every topic that isn't LLM hype related. You talk like a marketer of AI, someone that works adjacent to the industry with a dependency on these tools being bought.
> Please don't post insinuations about astroturfing, shilling, brigading, foreign agents, and the like. It degrades discussion and is usually mistaken. If you're worried about abuse, email hn@ycombinator.com and we'll look at the data.
You should delete this comment.
I was extremely skeptical at the beginning, and therefore critical of what was possible as my default stance. Despite all that, the latest iterations of cli agents which attach to LSPs and scan codebase context have been surprising me in a positive direction. I've given them tasks that require understanding the project structure and they've been able to do so. Therefore, for me my trajectory has been from skeptic to big proponent of the use, of course with all the caveats that at the end of the day, it is my code which will be pushed to prod. So I never went through the trough of disillusionment, but am arriving at productivity and find it great.
There are 3 parts of the process the AI agent can't do - the start, the middle and the end :) No, really, they need humans to identify tasks worth working on, then guide the model during development and providing iterative feedback, and in the end we incur the outcomes, good or bad. We are the consequence sinks, we take the costs and risks on ourselves. LLMs have no accountability.
I think that happened when gpt5 was released and pierced OpenAIs veil. While not a bad model, we found out exactly what Mr. Altman’s words are worth.
It feels that way to me, too—starting to feel closer to maturity. Like Mr. Saffron here, saying “go ham with the AI for prototyping, just communicate that as a demo/branch/video instead of a PR.”
It feels like people and projects are moving from a pure “get that slop out of here” attitude toward more nuance, more confidence articulating how to integrate the valuable stuff while excluding the lazy stuff.
[dead]