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

Thoughts on slowing the fuck down

mariozechner.at

Post reply on HN
Print view View original

501–505 of 505 posts

Re: Thoughts on slowing the fuck down

#501

Post by sevenseacat »

slhck wrote: post #411
chr15m wrote: post #379

Earlier quoted context omitted.

As well as pair programming with the AI, you can explicitly put those principles in AGENTS.md and the stochastic code generator will pay attention and be less verbose.

Except that when its system prompt is full of instructions, caveats, design principles, gotchas, architecture notes, memories from the past, and personal preferences, at some point it's going to just ignore them outright. Heck, Claude Code won't even use critical instructions from a 100-line CLAUDE.md file sometimes. So you still have to be extremely vigilant about noncompliance.

My CLAUDE.md is deliberately very short, and only includes very specific rules like "never list yourself as a co-author or committer in git commits". Claude will very regularly ignore this rule, apologize every time I tell it to fix it, update its memories, etc. and then an hour later do the exact same thing again.

Re: Thoughts on slowing the fuck down

#502

Post by sevenseacat »

skybrian wrote: post #87
badlibrarian wrote: post #5

I suppose everyone on HN reaches a certain point with these kind of thought pieces and I just reached mine. What are you building? Does the tool help or hurt? People answered this wrong in the Ruby era, they answered it wrong in the PHP era, they answered it wrong in the Lotus Notes and Visual BASIC era. After five or six cycles it does become a bit fatiguing. Use the tool sanely. Work at a pace where your understand…

People don't realize how much software engineering has improved. I remember when most teams didn't use version control, and if we did have it, it was crappy. Go through the Joel Test [1] and think about what it was like at companies where the answers to most of those questions was "no." [1] https://www.joelonsoftware.com/2000/08/09/the-joel-test-12-s...

The answer to those questions is _still_ 'no' at a lot of companies.

Re: Thoughts on slowing the fuck down

#503

Post by chr15m »

slhck wrote: post #411
chr15m wrote: post #379

Earlier quoted context omitted.

As well as pair programming with the AI, you can explicitly put those principles in AGENTS.md and the stochastic code generator will pay attention and be less verbose.

Except that when its system prompt is full of instructions, caveats, design principles, gotchas, architecture notes, memories from the past, and personal preferences, at some point it's going to just ignore them outright. Heck, Claude Code won't even use critical instructions from a 100-line CLAUDE.md file sometimes. So you still have to be extremely vigilant about noncompliance.

If your instructions are being ignored you may need a new model or harness.

Re: Thoughts on slowing the fuck down

#504

Post by alt227 »

alvivar wrote: post #482
alt227 wrote: post #449

Earlier quoted context omitted.

> There are too many people who need to discover for themselves what the limits of these AI models are when they push them far Why? Next week a new version of Claude and GPT will come out and the limits will change again. Are you really fully testing every new version of every LLM agant to see where its limits are? Those of us old enough to have seen this cycle before know its a fools game trying to keep up with deve…

Just curious, what have you seen before that was like AI?

ai in the sense of a new and immature techology which is constantly evolving and changing?

How about anything on the web? HTML boon in the 90s. Wordpress and PHP frameworks in the 00's, Javascript frontend shadow doms which require hydration in the 10's, node.js and javascript on the server etc etc.

All techonologies which were not worth jumping in during the initial boom becasue it changed so rapidly and it meant relearning concepts with every new release.

Its the same with ai, some prompt which gave amazing results last year might not now, and you need to be aware of what has changed and the better way to do it now. I prefer to jump in later when things are more mature and I can learn the most stable and liked way of doing something, instead of having to relearn the same thing multiple times as it changes.

Post reply on HN