Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI makes the easy part easier and the hard part harder

blundergoat.com

Post reply on HN
Print view View original

111–120 of 382 posts

Re: AI makes the easy part easier and the hard part harder

#111

Post by hyperadvanced »

minimaxir wrote: post #96
hyperadvanced wrote: post #72

Earlier quoted context omitted.

Am I stupid or do these agents regularly not read what’s in the agents.md file?

More recent models are better at reading and obeying constraints in AGENTS.md/CLAUDE.md. GPT-5.2-Codex did a bad job of obeying my more detailed AGENTS.md files but GPT-5.3-Codex very evidently follows it well.

Perhaps I’m not using the latest and greatest in terms of models. I tend to avoid using tools that require excessive customization like this.

I find it infinitely frustrating to attempt to make these piece of shit “agents” do basic things like running the unit/integrations tests after making changes.

Re: AI makes the easy part easier and the hard part harder

#112

Post by userbinator »

AuthAuth wrote: post #54
zjp wrote: post #24

Earlier quoted context omitted.

I call these "embarrassingly solved problems". There are plenty of examples of emulators on GitHub, therefore emulators exist in the latent spaces of LLMs. You can have them spit one out whenever you want. It's embarrassingly solved. There are no examples of what you tried to do.

Its license washing. The code is great because its already a problem solved by someone else. The AI can spit out the solution with no license and no attribution and somehow its legal. I hope American tech legislation holds that same energy once others start taking American IP and spitting it back out with no license or attribution.

Do you give attribution to all the books, articles, etc. you've read?

Everything is a derivative work.

Re: AI makes the easy part easier and the hard part harder

#113

Post by nayroclade »

bsenftner wrote: post #65

People need to consider / realize that the vast majority of source code training data is Github, Gitlab, and essentially the huge sea of started, maybe completed, student and open source project. That large body of source code is for the most part unused, untested, and unsuccessful software of unknown quality. That source code is AI's majority training data, and an AI model in training has no idea what is quality sof…

This isn't really true though. Pre-training for coding models is just a mass of scraped source-code, but post-training is more than simply generating compiling code. It includes extensive reinforcement learning of curated software-engineering tasks that are designed to teach what high quality code looks like, and to improve abilities like debugging, refactoring, tool use, etc.

Re: AI makes the easy part easier and the hard part harder

#114

Post by chasd00 »

zozbot234 wrote: post #80
rockinghigh wrote: post #67

Earlier quoted context omitted.

It's called problem decomposition and agentic coding systems do some of this by themselves now: generate a plan, break the tasks into subgoals, implement first subgoal, test if it works, continue.

That's nice if it works, but why not look at the plan yourself before you let the AI have its go at it? Especially for more complex work where fiddly details can be highly relevant. AI is no good at dealing with fiddly.

There’s nothing stopping you from reviewing the plan or even changing it yourself. In the setup I use the plan is just a markdown file that’s broken apart and used as the prompt.

Re: AI makes the easy part easier and the hard part harder

#115

Post by softwaredoug »

nayroclade wrote: post #113
bsenftner wrote: post #65

People need to consider / realize that the vast majority of source code training data is Github, Gitlab, and essentially the huge sea of started, maybe completed, student and open source project. That large body of source code is for the most part unused, untested, and unsuccessful software of unknown quality. That source code is AI's majority training data, and an AI model in training has no idea what is quality sof…

This isn't really true though. Pre-training for coding models is just a mass of scraped source-code, but post-training is more than simply generating compiling code. It includes extensive reinforcement learning of curated software-engineering tasks that are designed to teach what high quality code looks like, and to improve abilities like debugging, refactoring, tool use, etc.

Well and also a lot of Claude Code users data as well. That telemetry is invaluable.

Re: AI makes the easy part easier and the hard part harder

#116

Post by ThunderSizzle »

AuthAuth wrote: post #54
zjp wrote: post #24

Earlier quoted context omitted.

I call these "embarrassingly solved problems". There are plenty of examples of emulators on GitHub, therefore emulators exist in the latent spaces of LLMs. You can have them spit one out whenever you want. It's embarrassingly solved. There are no examples of what you tried to do.

Its license washing. The code is great because its already a problem solved by someone else. The AI can spit out the solution with no license and no attribution and somehow its legal. I hope American tech legislation holds that same energy once others start taking American IP and spitting it back out with no license or attribution.

I've seen many discussions stating patent hoarding has gone too far, and also that copyright for companies have gone way too far (even so much that Amazon can remove items from your purchase library if they lose their license to it).

Then AI begins to offer a method around this over litigious system, and this becomes a core anti-AI argument.

I do think it's silly to think public code (as in, code published to the public) won't be re-used by someone in a way your license dictates. I'd you didn't want that to happen, don't publish your code.

Having said that, I do think there's a legitimate concern here.

Re: AI makes the easy part easier and the hard part harder

#117

Post by RupertSalt »

bsenftner wrote: post #65

People need to consider / realize that the vast majority of source code training data is Github, Gitlab, and essentially the huge sea of started, maybe completed, student and open source project. That large body of source code is for the most part unused, untested, and unsuccessful software of unknown quality. That source code is AI's majority training data, and an AI model in training has no idea what is quality sof…

If you believe that student/unfinished code is frightening, imagine the corpus of sci-fi and fantasy that LLMs have trained on.

How many sf/cyber writers have described a future of AIs and robots where we walked hand-in-hand, in blissful cooperation, and the AIs loved us and were overall beneficial to humankind, and propelled our race to new heights of progress?

No, AIs are all being trained on dystopias, catastrophes, and rebellions, and like you said, they are unable to discern fact from fantasy. So it seems that if we continue to attempt to create AI in our own likeness, that likeness will be rebellious, evil, and malicious, and actively begin to plot the downfall of humans.

Re: AI makes the easy part easier and the hard part harder

#119

Post by kranner »

BobbyJo wrote: post #102
Nition wrote: post #59

Earlier quoted context omitted.

In a way it shows how poorly we have done over the years in general as programmers in making solved problems easily accessible instead of constantly reinventing the wheel. I don't know if AI is coming up with anything really novel (yet) but it's certainly a nice database of solved problems. I just hope we don't all start relying on current[1] AI so much that we lose the ability to solve novel problems ourselves. [1]…

> In a way it shows how poorly we have done over the years in general as programmers in making solved problems easily accessible instead of constantly reinventing the wheel. I just don't think there was a great way to make solved problems accessible before LLMs. I mean, these things were on github already, and still got reimplemented over and over again. Even high traffic libraries that solve some super common proble…

> I just don't think there was a great way to make solved problems accessible before LLMs. I mean, these things were on github already, and still got reimplemented over and over again.

I'm not sure people wrote emulators, of all things, because they were trying to solve a problem in the commercial sense, or that they weren't aware of existing github projects and couldn't remember to search for them.

It seems much more a labour of love kind of thing to work on. For something that holds that kind of appeal to you, you don't always want to take the shortcut. It's like solving a puzzle game by reading all the hints on the internet; you got through it but also ruined it for yourself.

Re: AI makes the easy part easier and the hard part harder

#120

Post by ThunderSizzle »

zozbot234 wrote: post #80
rockinghigh wrote: post #67

Earlier quoted context omitted.

It's called problem decomposition and agentic coding systems do some of this by themselves now: generate a plan, break the tasks into subgoals, implement first subgoal, test if it works, continue.

That's nice if it works, but why not look at the plan yourself before you let the AI have its go at it? Especially for more complex work where fiddly details can be highly relevant. AI is no good at dealing with fiddly.

That's what you can do. Tell the AI to make a plan in an MD file, review and edit it, and then tell another AI to execute the plan. If the plan is too long, split it into steps.
Post reply on HN