AI-first – We're just 6 months away from AGI
revontulet.dev
revontulet.dev
My personal take is that the whole chat based interface is a mistake. I have no better solution, but for anything beyond a halluciating search engine, it's not really the way we need to interact with AI.
In my mind we're 6 months away from one of the biggest crashes in tech history.
It's not just chat. But I really enjoy the Claude or chatgpt chat UI.
I'm not sure what your problem with that UI is tbh?
I don't think there will be any crash soon. I'm hoping for lower prices but I can't imagine not having these tools available anymore.
I'm also waiting for better and cheaper support for local development and inferencing.
I'm waiting left and right for a lot of smaller things. Like cheaper Claude, Hardware LLM chips, secure integration of my bank account and other personal data. More capacity for the beta projects of coding agents.
We don't need AGI to have increased productivity and benefits.
If it would be _that_ useless, half a billion people wouldn't use such tools every week (and this is not even counting Chinese users).
I think many hate LLMs because they realize that it starts to eat their bread and butter and are already taking their job and find it difficult to accept.
Investors are over-paying, because they are betting on companies that can be outpaced by others, but the technology itself is here to stay and grow, until it can do more and more tasks that are done by humans.
It doesn't mean that LLMs and other AI tools are going to run completely alone anytime soon (maybe with cars first ?), but eventually it could be.
What will be the world in 10 or 20 years from now ?
It's not a particularly efficient way to interact with the LLMs. Despite how people hate AI being integrated into everything, I do think that's the only way it can be done effectively. My take is that the AI/LLM isn't the product, it's just a feature of a product. Like when there's "AI" in the camera apps on your phone, or some sort of machine learning in medical imaging. To get the full benefit it needs to be a technology built into your product, not a product in it self.
The voice to text, text to voice is a good example. Voice to text is a feature of a transcription or video conference system. Text to voice is an accessibility feature. It doesn't matter that it's powered by an LLM.
If Claude, ChatGPT, whatever do become cheaper the market will crash. My thinking is: Prices go down (they will eventually), if that happens to fast, OpenAI will tank, their valuation is to high to support a $20 a month AI offering, or worse a world where ChatGPT isn't a product, but a feature. If OpenAI goes down the entire industry will have to re-align to a world where we now know that there's a limit to the value an LLM can command. Most of the current AI companies will be bought up at high discounts (as compared to their current valuations) and people are going to be laid off.
The personal assistant agent too.
They might overlap for me but not for everyone.
Nothing will crash. The big rich companies like Google, Microsoft already need it as part of their product strategy.
But about ai companies, I don't think their faith matters at all to end consumer and I'm pretty sure companies like character.ai do not run red numbers.
But this hype-storm just reminds me of the fever-dream blogs about the brave new world of the Internet back when hypertext became widely used in '93 or so (direct democracy, infinite commerce, etc, etc). Yes, of course, the brave new world came along, but it needed 3G and multi-touch screens as well and that was 15 years later and a whole different set of companies made money and ruled the world than those that bet on hypertext.
A few of my favorite tech authors and OSS contributors have gone full AI-parrot mode lately. Everything they post is either "how they use AI," "how amazing it is," or "how much it sucks." I think a bunch of us have had enough and just want to move on. But try saying anything even slightly critical about AI, or about how the big companies are forcing it into everything in the name of “progress,” and suddenly you're a Luddite.
I'm just glad more folks in the trenches are tired of the noise too and are starting to call it out.
This is very common. We need a name for this phenomenon. Any analogies/metaphors?
I generated (huh!) some suggestions and the funniest one seems to be:
> Just One More Prompt Syndrome (JOMPS)
But I assume there's a better one that describes or hints at the irony that describing precisely what a program should be doing _is_ programming.
The LLMs are sometimes frustrating, but over time you learn how to drive them properly, exactly like when you learn how to use the right keywords on Google.
Also, considering he is a free user, the author might be using one the default models ("auto" -> o4-mini or gpt-4o, rarely Claude Sonnet/Opus 3) in Cursor
-> Both of them are rather weak, especially with the automatic context splitting that Cursor does to save money
Claude Code + Opus 4 would likely yield better results here.
I wonder where you got the free user info.
In addition to that we've also seen that the way you prompt (amount of use of expert language), what context you provide through instructions, and tool use make a huge difference on the outcome.
At this point, if your coding experience with LLMs sucks I'd say there is an 80% chance that you're just doing it wrong.
Basically you are just adding to it with "You just need to use the latest model, anything before that is trash". Ignoring that the same was said just a few months prior when those models were cutting edge.
I'm not talking about other people's incorrect promises _and_ I mentioned a number of things in which proper usage today is different from what people were doing before.
> Basically you are just adding to it with "You just need to use the latest model, anything before that is trash".
That's not what I said. Don't put words into my mouth. I said that the older models are trash in comparison, not that they are trash. The older models require more work in prompting to get decent results.
> Ignoring that the same was said just a few months prior when those models were cutting edge.
You are conveniently ignoring the other parts I mentioned to make my claim. Here: "In addition to that we've also seen that the way you prompt (amount of use of expert language), what context you provide through instructions, and tool use make a huge difference on the outcome."
Alright? What you replied to and the context of this entire thread is about promises that have been made for a while now. In fact, we are approaching the point where we can safely talk about years of hype now. For reference, I am using the gpt-4 release as a significant ramp-up in the hype around LLMs.
> That's not what I said. Don't put words into my mouth.
That might not have been your intention. But, again, given the context of where you are replying it does read like. Even with the qualifier of "in comparison".
Like it or not, your comment does add extra qualifiers to the list.
Look, I am not saying that there isn't any progress being made here. I also agree that LLMs can be useful tools as part of a developer toolkit. What I personally don't agree with is that they can do the same job, even less so in real world scenarios. Even the latest models, including Gemini 2.5 and 03 struggle with moderately complex code base. And yes, the argument is always to let them work on small isolated bits of code. Or that if your requirements are tight enough they produce very good code. Which is entirely true, but I also envy the developers who work in structured environments where their code base is that clean and requirements that well-defined.
So, in my experience, coding with these models still sucks. Using them as interactive rubber duckies, replacements for some of the things I used to spend hours googling, debugging small snippets of code, etc. Sure, there they are very useful tools to me. But, to me, that is not coding with LLMs. That is having LLMs available as a tool whenever I need them.
You're getting there. The most valuable change is using software (like Cursor) that runs the models in agentic mode so they:
1. find the context themselves (with a basic document with context to point them in the right general direction for the current task).
2. can run commands and specifically tests.
3. iterate on their own output and changes to ensure you don't need to point out what they did wrong manually each time. Just let it find out itself, correct and keep moving until it is a good solution.
Think about how any human developer would approach an issue. For a lot of the 'moderately complex code bases', a developer new to it would also need a lot of pointing in the right direction, a lot of trying stuff out and then correcting themselves. Forget treating LLMs like one-shot magic solution givers, but instead as junior devs that you have to provide with all kinds of things to be successful.
Not to mention that with each iteration the amount of tokens needed goes up substantially. Having worked with LLM APIs and their pricing there simply is no way that Cursor is breaking even on that $20 or $40 per month if everyone uses it fully. Not even close. This very much hints at the costs being hidden right now, subsidized if you will, by VC money.
Also, once you have brought junior developers up to speed and guided them they are now slightly more capable developers who can more easily on board on future projects. With LLMs you need to effectively babysit them on each project again.
And there are a lot more caveats, prerequisites and moving targets involved that make the promise and reality for many people and companies not something they can actually be met.
And again, I am not discounting that there are specific areas where people see benefits from using LLMs in agentic from. But those areas are not as ubiquitous as the hype train leads us to believe. And to start using them you need to set up a lot more in the way of infrastructure and due process as well.
A late reply, but the point is that the old models are harder to get good results with, even in agentic mode. Therein lies the "massive difference in outputted code quality between models" I mentioned.
So using the newest models with the right approach quite easily leads to good results, which is what I meant with: "At this point, if your coding experience with LLMs sucks I'd say there is an 80% chance that you're just doing it wrong."
This as opposed to the original final claim made by the person I was replying to: "Just roll the dice correctly."
That implies that it is just about getting lucky with almost random output. It hasn't been that way for a long time, unless you really try to get the LLM to fail.
> With LLMs you need to effectively babysit them on each project again.
Only if you're doing it badly. Document and point the AI at documentation properly. Your onboarding process and documentation should not rely on people anyway. Write Once, Read Many.
There's a classical one: Sunk cost fallacy
One effective use of these tools is to get past the blank page dread. Let it write _something_ to get going, then turn it off and start working.
Recognizing where this cutoff should be can prevent "doom prompting"!
I suppose I could upload the image to a table to json website and provide copilot with the json but the point was to make things easier. In my mind there is nothing complicated about the structure of a table but if I ask copilot to merely extract the text from a row starting with some text it goes insane. Optical character recognition is an idea from 1870 and they had working implementations 50 years ago. I read a bunch of comments online about models getting progressively worse but my experience was over the span of a few weeks.
Would they be doctoring with the quality to make newer models look even better?
Making a misleading title in order to bait responses would be clickbait. People responding to a title that assertively states something is not some amazing gotcha.
Although in this case it must be what the sibling comment said about HN stripping the smiley.
To be clear, bullish has a few meanings. But they all boil down to assertively communicating. From your comment I gather you might think it's etymology traces back to "bullying", but it actual relates back to "bull".
I don't think anyone expected it to make all books obsolete, but you'd struggle to buy an encyclopaedia that isn't for kids these days.
We have long term memory and short term.
Context is short term.
The still long and expensive training phase embeds the long term memory.
But do I really have to say colloquial?
One of the problems programmers have is loading a problem into working memory. It can take an hour. An interruption, a phone call, or a meeting can mean that you have to start over (or, if not completely over, you still have to redo part of it). This is a standard programmer complaint about interruptions.
It's interesting that LLMs may have a similar issue.
Right now inference doesn't cascade into training.
In biology, inference and training are not so decoupled.
Millions of people use LLMs daily.
Veo 3 just came out.
People get fired due to LLMs and GenAI replacing them.
Google uses ai code for years in the background.
Playing around with cursor for a few days is way to naive to be able to determine the current situation with coding and ai
Serious question: Are they? Or are they being fired and AI is used at the excuse?
We saw Klarna layoff customer services staff and that didn't work. LLMs couldn't do their job. Some programmers are being fired, but my feeling is that generative AI is more of a convenient excuse.
Is anyone actually being fired because an LLM did their job better, or where their job already in danger and AI just gave the companies an easy way out?