They can think and reason better than most humans. Most problems they're pointed at are not in their training set, but in certain ways they resemble things that are—maybe there are a few different resemblances to different problems in its training set—so it's able to pull these disparate similarities together and apply the patterns it finds to come up with a solution. Much like human brains do.
What? This is a massive misunderstanding. It’s easy to get truly novel ideas from LLMs, unless your definition of “new things” is so strict that no human can do so either.
The training set is about patterns, not facts or specific configurations. Yes, it’s possible to extract (some) of the training set verbatim, but that doesn’t mean it’s all you can do.
>unless your definition of “new things” is so strict that no human can do so either.
Humans rarely think of new things. We're a weak hivemind species. One or two individuals figure something out, and the rest of the troop of monkeys imitates. Brains are too fuel hungry for every brain to be innovating, "innovate and copy to the other brains" is the norm.
It does not think at all. It vibes based on its training and any additional bolted on constraints. It is a quite simple automation that only works by huge amount of existing data.
Modern man has grown quite dumb. He only seems to be able to "invent" by massive scaling things that are decades or centuries old..
Electricity runs from simple batteries (600 BCE) to today’s power grids.
RF was predicted but not demonstrated by Maxwell in the 1860’s. His work built on Faraday’s (1840’s) and Coulomb’s (1780’s). Coulomb built on Franklin and Newton, among others. Or do you mean Marconi and Tesla, who merely implemented what Maxwell predicted?
The same is true for lasers and transistors but it’s tedious. There was no single “back in the day people invented things from whole cloth” moment.
I would put it differently. Those inventions came from humans interacting with the physical world.
When LLMs were first introduced, they didn't have much of a feedback loop. They wrote code, but they couldn't compile it. Not surprisingly, the code had bugs.
Now, they run with harnesses that allow them to compile the code, and react to the issues they observe. They can fix their own bugs and solve problems that they create, just like humans.
Give an agent access to the physical world, and it seems highly likely that they will be able to "invent" things based on feedback they receive while working towards goals.
Of course, there are some well-known limitations of LLMs, one of the biggest being that they're pretrained. So there may be some things where they're not as good. Just like how some humans aren't as good at certain tasks, depending on their genetics and/or how they've been trained.
Those are not merely scaling. I can get “build upon other works”, but there’s a lot of scientific insights needed for observing and modeling a phenomena. It may even requires a boost of creativity to theorize an effect based on that model and then make it possible in an experiment.
Individual brilliance has always been rare. If I booted your ass back 1000 years you'd have seen the same shade of grey in the pig farmers and peasants along with most of the nobles and kings. It turns out that in general the world is a shade of grey.
Hell, looking back at the brilliant individuals, they had many, many grey areas.
I was pointing out two things: first, your understanding of LLM capabilities is very outdated; and second, that in this respect, you're behaving much like an LLM with a training cutoff.
That further touches on the idea that the differences between you and an LLM may not be as large as you imagine. In particular, "cobbling something together if it's in their training set" is pretty much what all humans do.
This reminds me of the Go champion who announced he was giving up the game after a computer beat him.
It’s as if a runner were to give up running when beaten by a horse or a car. It suggests they may have had unexamined and perhaps somewhat strange reasons for doing the activity in the first place.
People have difficulty accepting just how significant their limitations actually are. We design our world to hide those limitations. As an example, it would be easy to make computer games that are unwinnable by humans because of our slow reaction times, low speed in general, and our cognitive limitations. But no-one makes such games, because few people would want to play them for very long.
The “terrible cost” in this specific case seems to be related to discovering that we were fooling ourselves about how good we were at software development.
I'm not giving up the career. And I certainly don't feel left behind: I'm good at programming, and I still have a substantial edge over non-programmers here in meaningfully using agents (as of June 2026). The terrible cost refers to sucking all the joy of the process as the mechanized activity makes the actual intellectual part of the work redundant, if you can understand.
Why is it bittersweet? Carpenters probably didn't cry when their tools improved.
It will be bittersweet when there's no human needed at the wheel but IMHO we are far, far from that. These models/agents are just mimicking human text and need guidance because they often get lost or stuck.
Carpenters would have cried if all their work was reduced to shoving the logs into CNC machines.
Yes there is still human input but it requires comparatively no skill or depth and it gets easier by the month. If I were lobotimized today I'd still be able to function as half-assed architect to AIs anyway.
When was the last time you read fighting distractions/getting "in the zone"/complaint about open space offices thread or comment? They used to be a weekly feature on HN frontpage.
> Yes there is still human input but it requires comparatively no skill or depth and it gets easier by the month. If I were lobotimized today I'd still be able to function as half-assed architect to AIs anyway.
Hard doubt, software engineering is so much more than just literal coding and typing. At least for many of us, the coding/typing part is the easy stuff, everything around that is where the actual engineering happens. If I were lobotomized, maybe I'd get ~10% done today as the day before, if I'm lucky. Even with my full mental capabilities, the agents end up on wild goose-chases unless I'm very specific with what I want, and even sometimes ignoring things if they're too complicated/takes too long, so a bit of thinking is still required to get the right prompts.
And considering how subjective programming is, since it's a creative endeavour after all, I'm not that worried somehow all programmers will be unemployed in just some years.
> When was the last time
Frequency of something doesn't tell you how big of an issue something is, for all we know, HN community (or even the moderators) could have been tired of all the circular conversations where nothing new is being said, and downvote it. Doesn't really tell us much.
Coding is literally writing code, instructions in plain text that control the behavior of computer. That implies knowing which instructions to write.
But creating software is much more than that. Just like writing an essay involves more than just typing words. Other activities include: Architecture, Requirements analysis, Debugging, Testing, Integration,…
Use whatever labels you want, apply charitable reading and I'm sure even you could understand what I mean here. Clearly there are at least two sorts of tasks (or used to anyways) in "software engineering" as a whole, one more mechanical and one more about thinking.
But it's not like "shoving the logs into CNC machines". You have to understand what they are doing and point them into the right direction. LLMs very often lack common sense once you move out of easy things.
I love programming CNC machines; I am a terrible carpenter. Someone still has to tell the LLMs what to build, specify design constraints and goals, etc
I think carpenters might cry if a company went around shoving every single piece of carpentry they could find into a machine, and then when you press a button on that machine, a chair comes out, and then they go around saying that this machine will replace carpenters forever, and they made this machine with no help from other carpenters, and furniture makers all went "who needs carpenters anymore, lets just use the chair machine"
Software isn't solved. 'Coding' is, according to the people of Claude.
Coding (programming) is a tedious and expensive part of software engineering. There's other parts AI isn't doing, such as understanding and refining requirements, and delivery + accountability.
> Coding (programming) is a tedious and expensive part of software engineering.
Why is that? Coding, for me, is kinda relaxing, and the fun part of developing software. Gathering requirements, especially in a corporate settings, is the tedious part and the most time consuming.