Hacker News Reader

Comment by allovertheworld | Hacker News Reader

I found you have to start entirely new sessions once you reach a “flagged” hardcoded response like this. You cant escape it once it is poisoned

View on HN

This is almost certainly the issue. It's very unintuitive for users, but LLMs behave much better when you clear the context often. I run /clear every third message or so with Claude Code to avoid context rot. Anthropic describes this a bit with their best practices guide [0].

[0] https://www.anthropic.com/engineering/claude-code-best-pract...

Can we agree that this is no longer programming?

I don't know what it is, but trying to coax my goddamn tooling into doing what I want is not why I got into this field.

(I agree we shouldn't call it programming)

Uhm – isn't "coax my goddamn tooling into doing what I want" basically all we did pre-LLMs anyway?

How often did you IDE or editor refuse to do something it was generally capable of because it deemed the operation too frivolous in a context?

I think this is what binary programmers/hole punching said about programming languages

This'd be a valid analogy if all compiled / interpreted languages were like INTERCAL and eg. refused to compile / execute programs that were insufficiently polite, or if the runtime wouldn't print out strings that it "felt" were too silly.

Now there's an idea for an esoteric language.

It depends from which vantage point you look at it. The person directing the company, let's imagine it was Bill Gates instructing that the code should be bug free, but its very opinionated about what a bug is at Microsoft.

> I don't know what it is, but trying to coax my goddamn tooling into doing what I want is not why I got into this field.

I can understand that, but as long as the tooling is still faster than doing it manually, that's the world we live in. Slower ways to 'craft' software are a hobby, not a profession. (I'm glad I'm in it for building stuff, not for coding - I love the productivity gains).

This isn't likely to be a hardcoded type of classified response. I think this response is literally "you offended the model persona's sensibilities." But, yes after the first denial the models will double down.

Reply on news.ycombinator.com