Hacker Newsnew | past | comments | ask | show | jobs | submit | quantumHazer's commentslogin

> Fable didn't even need to debug it; I believe essentially all the assembly worked correctly on the first try.

so you don't know if it works properly or not


Fable wrote a pretty decent test suite covering typical Prolog programs, including things with non-trivial execution patterns like "append" that do complex backtracking on multiple branches. And I've run a modest number of test queries by hand, trying a few things. It's entirely possible that there's a bug there somewhere. But it's better than I would have done on the first try, implementing a Prolog interpreter in assembly. And I've worked on actual production compilers a few times.

I am honestly not happy about the way that models can now just take what should have be a fun multi-weekend project and knock out in a couple of hours. But I'm not going to pretend that Fable is stupid, or that it did a bad job on any of the test projects I gave it. It struggles more on big, messy real-world code bases, absolutely.


It's also conveniently forgetting that AI was trained on the whole internet, a corpus of code including every garbage solution in SO wrong answers or random junior blogs. I didn't see any claim AI could select "best practices" from their sources, so putting AI to correct another AI has high chances of statistically selecting the same wrong solution, because that was just a common thread in those old posts they used as reference.


I think you're pretty wrong here, but I'm no expert so anyone who wants to correct me please do so.

As I understand it, there's post-training step that does exactly that, they get real developers to select good and bad code and help the AI figure that out. AFAIK it's the reason why Claude was able to leap-frog OpenAI in coding last year, they did this refinement step a lot better.

And now they've taught the older models this, they can use existing models to bootstrap the newer models without humans involved, and bring humans in to just assess the trickier stuff.


Claude's coding efficiency is largely due to their control plane governing the actual execution of the model - rules, evaluators and basically looping over results. Not related to the model training or fine tuning.


The models were already much better before Claude Code came out, let's not rewrite history here.


Probably llm slop


if its useful and entertaining i dont think many ppl care especially evident that you cannot be certain with your accusations


How is it useful? I'm no expert in combustion engines, so I'd hesitate to rely on a slop graphic like this to learn about them as I doubt the creator has any idea about them either.

It's like a friend - he was like "Look at this awesome set of categorized interactive animations from claude to learn geometry!".

And like 60% of the animations were technically just wrong. Very pretty though, and the effects were cute... I guess...?


why are you so invested in telling ppl in what they are allowed to find useful/entertaining?

especially after admitting you dont nkow anything about combustion engines?


Where did I say what people are allowed to find entertaining? It would be insane to tell someone what they should find entertaining, that's the dumbest thing I've ever heard.

As for useful, I'm genuinely trying to understand HOW you find something useful that may or may not have zero accuracy to the thing in which it's portending to showcase?

When I don't know about something - I would find someone who's really knowledgeable (or a resource that's vetted), and learn from that obviously. I don't want to learn something completely unverified and made up? That makes no logical sense.

Just like my friend's slop graphics. I imagine learning geometry from that and fundamentally locking in incorrect knowledge when there are many thousands of resources that would teach me correctly - this makes no sense right? Would you prefer to learn from the pretty graphics that are teaching you nothing?


> it would be insane to tell someone what they should find entertaining, that's the dumbest thing I've ever heard.

thats what you wrre doing above read it again if it wasnt obvious


Here I will quote what I said to you - it's super explicit and obvious I'm asking how one would find it useful.

"How is it useful? I'm no expert in combustion engines, so I'd hesitate to rely on a slop graphic like this to learn about them as I doubt the creator has any idea about them either."

Notice I say "How is it useful?" and "I'd hesitate to rely on a slop graphic like this to learn".

I think I'm being very obvious about the fact that I can't understand how one would find it "useful" to "learn". Absolutely NOWHERE did I say you can't be entertained by it.


please dont use AI to generate comments.

https://news.ycombinator.com/newsguidelines.html


I'm certain.


> There are many ways to build for the web, both with and without AI. If AI is a part of your workflow, we think this tool will help make it even more productive. And if it isn’t, that’s OK too.

Crazy thing to say in 2026 where if you write code and not delegate every bit to an agent you're considered a noob by some people.


The opposite was considered two years ago a crazy thing to say. I'm glad this changed and people using ai don't have to hide in the closet anymore for doing so.


When have people using AI to develop ever had to hide in any closet? It's the most public "new hotness" ever.


That’s the point I think. Remember the controversy when Github Copilot came out? Not because where it got its data from, but because people didn’t feel like they wrote code anymore, they just tabbed the autocompleted snippets and was finished with a task in much shorter time.


It wasn't nearly as good 2 years ago.


Which part makes it crazy? The fact that they felt the need to add this disclaimer to try to assuage the fanatical anti-AI contingent?

Like this is specifically a tool for AI-augmented development, and they had to add this "but also, thoughts and prayers for you non-AI people" is incredibly weird, but not in the way you seem to think it is.


also, there is multiple models called codex or that have codex as a "suffix" eg. "gpt-5.3-codex"


> not using GenAI could mean project won't be competitive in current landscape

why? this is false in my opinion, iterating fast is not a good indicator of quality nor competitiveness


iterating fast over quality (e.g. refactoring, tests coverage, benchmarks, documentation, trying new nontrivial ideas) is a good indicator of quality.


you can’t iterate fast over quality though. it takes patience and expertise, not a bloated repo like this.

every example you mentioned is not something you should delegate to LLMs, unless quick prototyping


> every example you mentioned is not something you should delegate to LLMs, unless quick prototyping

it works very well for me, llm with guidance produces good quality code.


the author is an employee of Cursor/Anysphere. i think this should've been declared at the start of the article. it's always like this.


Not unpopular at all. I think it's one of the best Ferrari ever made.


Or maybe, hear me out, we don't need any of this ""agent"" first shiny thingy


I wouldn’t be surprised if this is undisclosed PR from Anthropic


I'd be very surprised if it wasn't. Everything about that company turns me off. I've run across countless YouTube videos that are clearly Anthropic PR pretending to be real videos by regular people just trying it out and discovering how good Claude is. I'll stick with Gemini.


Why are we commenting the Claude subreddit?

1) it’s not impartial

2) it’s useless hype commentary

3) it’s literally astroturfing at this point


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: