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

6 weeks of Claude Code

blog.puzzmo.com

Post reply on HN
Print view View original

541–550 of 603 posts

Re: 6 weeks of Claude Code

#541

Post by bredren »

ifwinterco wrote: post #533
michaelrpeskin wrote: post #478

Earlier quoted context omitted.

Yes, this! A couple of week ago, I had a little down time and thought about a new algorithm I wanted to implement. In my head it seemed simple enough that 1) I thought the solution was already known, and 2) it would be fairly easy to write. So I asked Claude to "write me a python function that does Foo". I spent a whole morning going back and forth getting crap and nothing at all like what I wanted. I don't know what…

The LLM will never admit it doesn't have a clue what's going on, but over time you develop a sense of when it's onto something and when it's trapped in a loop of plausible sounding nonsense Edit: Also, it's funny how often you can get it to improve its output by just saying "this looks kind of bad for x reason, there must be a way to make it better"

I have experimented with instructing CC to doubt itself greatly and presume it is not validating anything properly.

It caused it to throw out good ideas for validation and working code.

I want to believe there is some sweet spot.

The constant “Aha!” type responses followed by self validating prose that the answer is at hand or within reach can be intoxicating and can not be trusted.

The product is also seemingly in constant flux of tuning, where some sessions result in great progress, others the AI seems as if it is deliberately trying to steer you into traffic.

Anthropic is alluded toward this being the result of load. They mentioned in their memo about new limits for Max users that abuse of the subscription levels resulted in ~subpar product experiences. It’s possible they meant response times and the overloaded 500 responses or lower than normal TPS, but there are many anecdotal accounts of CC suddenly having a bad day from “longtime” users, including myself.

I don’t understand how load would impact the actual model’s performance.

It seems like only load based impacts on individual session context would result in degraded outputs. But I know nothing of serving LLM at scale.

Can anyone explain how high load might result in an unchanged product performing objectively worse?

Re: 6 weeks of Claude Code

#542

Post by jmtame »

azuanrb wrote: post #485
jmtame wrote: post #295

Earlier quoted context omitted.

I only use 2 MCP servers, and those are context7 and perplexity. For things like updated docs, I have it ask context7. For the more difficult technical tasks where I think it's going to stumble, I'll instruct Claude Code to ask perplexity and that usually resolves it. Or at least it'll surface up to me in our conversation so that we both are learning something new at that point. For some new stuff I'm working on, I u…

Thanks, I’ll check out Perplexity. We seem to be using a similar stack. I’m also on Rails 8 with Stimulus, Hotwire, esbuild, and Tailwind. Playwright MCP has been a big help for frontend work. It gives the agent faster feedback when debugging UI issues. It handles responsive design too, so you can test both desktop and mobile views. Not sure if you know this, but Claude Code also works with screenshots. In some cases…

I didn’t realize screenshots worked until a few days in, that was a great discovery. And recently learned you can directly paste in copied screenshots using ctrl+v (instead of cmd+v on a Mac).

Re: 6 weeks of Claude Code

#543

Post by awesome_dude »

leptons wrote: post #418
aryonoco wrote: post #383

Earlier quoted context omitted.

No one is skeptical of compilers?! I guess you haven’t met many old fashioned C systems programmers, who go out of their way to disable compiler optimisations as much as they can because “it just produces garbage”. Every generation, we seem to add a level of abstraction conceding because for most of us, it enhances productivity. And every generation, there is a crowd who rails against the new abstraction, mostly unaw…

Abstractions and compilers are deterministic, no matter if a neckbeard is cranky about the results. LLMs are not deterministic, they are a guessing game. An LLM is not an abstraction, it's a distraction. If you can't tell the difference, then maybe you should lay off the "AI" slop.

I've been thinking about this - you're right that LLMs are not going to be deterministic (AIUI) when it comes to producing code to solve a problem.

BUT neither are humans, if you give two different humans the same task, then, unless they copy one another, then you will get two different results.

Further, as those humans evolve through their career, the code that they produce will also change.

Now, I do want to point out that I'm very much still at the "LLMs are an aid, not the full answer.. yet" point, but a lot of the argument against them seems to be (rapidly) coming to the point where it's no longer valid (AI slop and all).

Re: 6 weeks of Claude Code

#544

Post by aswegs8 »

makingstuffs wrote: post #525
willsmith72 wrote: post #503

Earlier quoted context omitted.

Why does it go against your morality? Sounds like a totally rational business decision, only affecting a sub-par partner

I know what you mean it just feels a bit inhumane to me. Sort of like defining a value for a living being and then determining that they fell beneath said value.

Yeah well, at some point some one higher up without scruples about such moral issues will make that decision for you...

Re: 6 weeks of Claude Code

#545

Post by exfalso »

fourthark wrote: post #526
exfalso wrote: post #396

Earlier quoted context omitted.

Hmm not my experience. I've been aggressively trying to use both Cursor and Claude Code. I've done maybe 20-30 attempts with Code at different projects, a couple of them personal small projects. All of them resulted in sub-par results, essentially unusable. I tried to use it for Python, Rust and Bash. I also tried to use it for crawling and organizing information. I also tried to use it as a debugging buddy. All of t…

You're probably in an obscure niche domain, or asking it to do something creative. Try like upgrading JS package dependencies, or translating between languages, limited tedious things, and you will be surprised how much better it does.

Hmmmm.. I am working in a niche domain (Confidential Computing) and the work is fairly creative, although I wouldn't say I asked it domain-specific things. I didn't ask it to come up with encryption schemes or security protocols, I learned very quickly that it cannot even start on those problems. "Design discussions" were just sycophantic affirmations of whatever I wrote. What I mostly tried were "add this function" or "refactor this based on XY" or "analyze this piece of code for race conditions".

(Un?)fortunately my work doesn't involve a lot of "drone coding". With personal projects I let it do whatever it wanted including picking the language and libraries. With one of them it ended up so confused with the Redis API(!!!) that it kept going back and forth between different versions to "fix" the issues until it literally removed the functionality it was supposed to add. Problem solved, eh?

Re: 6 weeks of Claude Code

#546

Post by kundi »

globular-toast wrote: post #453
yubblegum wrote: post #292

Earlier quoted context omitted.

> I'm hearing from Senior devs all over thought, that Junior developers are just garbage at it. They product slow, insecure, or just outright awful code with it, and then they PR the code they don't even understand. If this is the case then we better have full AI generated code within the next 10 years since those "juniors" will remain atrophied juniors forever and the old timers will be checking in with the big cloc…

You can't atrophy if you never grew in the first place. The juniors will be stunted. It's the seniors who will become atrophied. As for whether it's a mistake, isn't that just the way of things these days? The current world is about extracting as much as you can while you're still here. Look around. Nobody is building for the future. There are a few niche groups that talk about it, but nobody is really doing it. It's…

Sad, but very true observation

Re: 6 weeks of Claude Code

#547

Post by yencabulator »

base698 wrote: post #498
floppyd wrote: post #124

Anybody had similarly good experience with Gemini CLI? I'm only a hobbyist coder, so paying for Claude feels silly when Gemini is free (at least for now), but so far I've only used it inside Cline-like extensions

I’ve used both. Claude more extensively. I’ve had good results with Gemini too, however it seems easier to get stuck in a loop. Happens with Claude too but not quite as frequent. By loop I mean you tell it no don’t implement this service, look at this file instead and mimic that and instead it does what it did before.

I regularly have Claude Code in a loop where it can't figure out Typescript types and uses unsafe `as` kludges, even when CLAUDE.md tells it not to. A couple of prompts later, if it encounters any error in that same region, it again rips out the typesafe code and replaces it with an `as`.

My observation has been this: if you push a(/any) current-day LLM too close to the edge of its abilities, it goes "insane". Hallucinations start happening everywhere, it stops ignoring previous knowledge, etc. The best way out is to end the session, maybe do some manual work to get to a good state, perhaps update the specs, and start with a fresh context. Using "strong words" or prompting more is of no consequence, the LLM will produce essentially gibberish until reset. Sometimes using a more expensive model temporarily gets around whatever is triggering the stupidity.

Re: 6 weeks of Claude Code

#548

Post by antonvs »

radley wrote: post #304
dimal wrote: post #297

Earlier quoted context omitted.

Technically you’re not vibe coding. You’re using AI to do software engineering. Vibe coding is specifically the process of having AI produce code and plowing ahead without understanding it. I know I’m being pedantic, but people mean very different things when they talk about this stuff, and I don’t think any credence should be given to vibe coding.

Both are vibe coding. The term was coined by Andrej Karpathy, a computer scientist who served as the director of artificial intelligence at Tesla. Maybe you're thinking of slop coding ?

Here's the full original definition from Karpathy:[*]:

> There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper so I barely even touch the keyboard. I ask for the dumbest things like "decrease the padding on the sidebar by half" because I'm too lazy to find it. I "Accept All" always, I don't read the diffs anymore. When I get error messages I just copy paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension, I'd have to really read through it for a while. Sometimes the LLMs can't fix a bug so I just work around it or ask for random changes until it goes away. It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or webapp, but it's not really coding - I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.

Some of the key points here being "forget that the code even exists," "'Accept All' always," "don't read the diffs," and "The code grows beyond my usual comprehension."

Doing software engineering using AI assistance is not vibe coding, by Karpathy's definition.

[*] https://x.com/karpathy/status/1886192184808149383

Re: 6 weeks of Claude Code

#549

Post by yencabulator »

strange_quark wrote: post #235
nrclark wrote: post #223

Does Claude Code use a different model then Claude.ai? Because Sonnet 4 and Opus 4 routinely get things wrong for me. Both of them have sent me on wild goose chases, where they confidently claimed "X is happening" about my code but were 100% wrong. They also hallucinated APIs, and just got a lot of details wrong in general. The problem-space I was exploring was libusb and Python, and I used ChatGPT and also Claude.ai…

Pretty sure it’s the same model. The reason that claude code is “good” is because it can run tests, compile the code, run a linter, etc. If you actually pay attention to what it’s doing, at least in my experience, it constantly fucks up, but can sort of correct itself by taking feedback from outside tools. Eventually it proclaims “Perfect!” (which annoys me to no end), and spits out code that at least looks like it s…

> Eventually it proclaims “Perfect!” (which annoys me to no end),

This has done wonders for me:

  # User interaction
  - Avoid sycophancy.
  - Do what has been asked; nothing more, nothing less.
  - If you are asked to make a change, summarize the change but do not explain its benefits.
  - Be concise in phrasing but not to the point of omission.

Re: 6 weeks of Claude Code

#550

Post by ako »

ModernMech wrote: post #520
ako wrote: post #494

Earlier quoted context omitted.

I don't agree, to me switching from hand coded source code to ai coded source code is like going from a hand-saw to an electric-saw for your woodworking projects. In the end you still have to know woodworking, but you experiment much more, so you learn more. Or maybe it's more like going from analog photography to digital photography. Whatever it is, you get more programming done. Just like when you go from assembly…

> to me Key words there. To you, it's a electric saw because you already know how to program, and that's the other person's point; it doesn't necessarily empower people to build software. You? Yes. Generally though when you hand the public an electric saw and say "have at it, build stuff" you end up with a lot of lost appendages. Sadly, in this case the "lost appendages" are going to be man-decades of time spent undo…

Being able to do it quicker and cheaper will often ensure more people will learn the basics. Electrical tools open up woodworking to more people, same with digital photography, more people take the effort to learn the basics. There will also be many more people making rubbish, but is that really a problem?

With ai it’s cheap and fast for a professional to ask the AI: what does this rubbish software do, and can you create me a more robust version following these guidelines.

Post reply on HN