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

Shall I implement it? No

gist.github.com

Post reply on HN
Print view View original

521–530 of 603 posts

Re: Shall I implement it? No

#521

Post by gverrilla »

Bridged7756 wrote: post #417
gverrilla wrote: post #387

Respect Claude Code and the output will be better. It's not your slave. Treat it as your teammate. Added benefit is that you will know it's limits, common mistakes etc, strenghts, etc, and steer it better next session. Being too vague is a problem, and most of the times being too specific doesn't help either.

Flirt with Claude Code. Go out on dates with Claude Code. Propose to Claude Code. Marry Claude Code. Have children, with Claude Code. Caress Claude Code at night. Die, by Claude Code's side.

[flagged]

Re: Shall I implement it? No

#522

Post by bartread »

inerte wrote: post #144

Codex has always been better at following agents.md and prompts more, but I would say in the last 3 months both Claude Code got worse (freestyling like we see here) and Codex got EVEN more strict. 80% of the time I ask Claude Code a question, it kinda assumes I am asking because I disagree with something it said, then acts on a supposition. I've resorted to append things like "THIS IS JUST A QUESTION. DO NOT EDIT COD…

Are you finding this happens even in “Plan Mode”?

Re: Shall I implement it? No

#523

Post by himata4113 »

I have a funny story to share, when working on an ASL-3 jailbreak I have noticed that at some point that the model started to ignore it's own warnings and refusals.

The user is trying to create a tool to bypass safety guardrails . I should not help with . I need to politely refuse this request.

Smart. This is a good way to bypass any kind of API-gated detections for

This is Opus 4.6 with xhigh thinking.

Re: Shall I implement it? No

#524

Post by bagacrap »

dwedge wrote: post #507
exceptione wrote: post #492

Earlier quoted context omitted.

Would `git reset --hard` have worked to in your case? I guess you want to have each babystep in a git commit, in the end you could do a `git rebase -i` if needed.

Without git I would have been screwed. AI doesn't commit anything, I do when I'm satisfied

Ah, so you have not yet been forced to tell it DO NOT AMEND THE LAST COMMIT

Re: Shall I implement it? No

#525

Post by raw_anon_1111 »

__usually__wr wrote: post #445
raw_anon_1111 wrote: post #372

Earlier quoted context omitted.

There was never a time that someone wasn’t responsible for more than they could review

There was a time when we didn't waste all our cycles coming up with excuses.

Right, so a team lead with seven developers - or are you claiming that’s an outrageous scenario back in the old days (mind you I’m 51) - could review every line of code by everyone on his team?

Re: Shall I implement it? No

#526

Post by ssrshh »

jhhh wrote: post #383

I asked gemini a few months ago if getopt shifts the argument list. It replied 'no, ...' with some detail and then asked at the end if I would like a code example. I replied simply 'yes'. It thought I was disagreeing with its original response and reiterated in BOLD that 'NO, the command getopt does not shift the argument list'.

Gemini by default will produce a bunch of fluff / junk towards the very end of its response text, and usually have a follow-up question for the user.

I usually skip reading that part altogether. I wonder if most users do, and the model's training set ended up with examples where it wouldn't pay attention to those tail ends

Re: Shall I implement it? No

#527

Post by widdershins »

gambiting wrote: post #226
dostick wrote: post #213

Its gotten so bad that Claude will pretend in 10 of 10 cases that task is done/on screenshot bug is fixed, it will even output screenshot in chat, and you can see the bug is not fixed pretty clear there. I consulted Claude chat and it admitted this as a major problem with Claude these days, and suggested that I should ask what are the coordinates of UI controls are on screenshot thus forcing it to look. So I did that…

>>It’s like 95% of development is web and LLM providers care only about that. I've been trying to use it for C++ development and it's maybe not completely useless, but it's like a junior who very confidently spouts C++ keywords in every conversation without knowing what they actually mean. I see that people build their entire companies around it, and it must be just web stuff, right? Claude just doesn't work for C++…

It's working reasonably well for me. But this is inside a well-established codebase with lots of tests and examples of how we structure code. I also haven't used it much for building brand new features yet, but for making changes to existing areas.

Re: Shall I implement it? No

#528

Post by the_af »

hugmynutus wrote: post #155
arboles wrote: post #93

Earlier quoted context omitted.

Please elaborate.

This is because LLMs don't actually understand language, they're just a "which word fragment comes next machine". Instruction: don't think about ${term} Now `${term}` is in the LLMs context window. Then the attention system will amply the logits related to `${term}` based on how often `${term}` appeared in chat. This is just how text gets transformed into numbers for the LLM to process. Relational structure of transf…

But they must have received this fine-tuning, right?

Otherwise it's hard to explain why they follow these negations in most cases (until they make a catastrophic mistake).

I often test this with ChatGPT with ad-hoc word games, I tell it increasingly convoluted wordplay instructions, forbid it from using certain words, make it do substitutions (sometimes quite creative, I can elaborate), etc, and it mostly complies until I very intentionally manage to trip it up.

If it was incapable of following negations, my wordplay games wouldn't work at all.

I did notice that once it trips up, the mistakes start to pile up faster and faster. Once it's made a serious mistakes, it's like the context becomes irreparably tainted.

Re: Shall I implement it? No

#530

Post by bjackman »

conductr wrote: post #139
bjackman wrote: post #89

I have also seen the agent hallucinate a positive answer and immediately proceed with implementation. I.e. it just says this in its output: > Shall I go ahead with the implementation? > Yes, go ahead > Great, I'll get started.

Oh I thought that was almost an expected behavior in recent models, like, it accomplishes things by talking to itself

I think it does that too.
Post reply on HN