Hacker News Reader

Hacker News

Comment by a1o | Hacker News Reader

Full thread

a1o

> What's modeled vs. measured

> Honesty section:

I wouldn’t mind an human written README

View on HN

> Honesty section:

This one gets to me. Not only is it useless (paragraph would have the same meaning if it was removed) but wdym "honesty section," Claude? The whole readme should be honest! One of the many causes of psychological damage when reading AI-generated technical writing is the reminder that one of its tics is that declares when it's being honest or real, casting doubt on the rest.

Don't you ever add a few dishonest paragraphs to your docs to spice things up?

The other day the system prompts for Claude (chat not code) hit the frontpage and they included a bit that something along the lines of "Claude should avoid saying honestly, because Claude is always honest". So how come that these Claudisms still are so frequent in LLM output? Do the system prompts just not work? Don't the postprocess the output to deal with such policy violations?

Some other day here HN someone posted a blog post that was like a long nonsense text of claudisms like load bearing smoking gun and other things, it was beautiful. I wanted to have a link for it to show to others but it disappeared when I got back to look for it and I lost the link. :/

https://aloutfi.com/writing/load-bearing-smoking-gun

I love it.

Imagine instructing a fentanyl addict not to consume any fentanyl. That's how system prompts work sometimes.

Something somewhere in the training pipe has perturbed the LLM in a way that makes it say "honestly" all the time. It's a weasel word. It's a habit. It's an addiction. It's borderline involuntary. Even changing the system prompt only makes it 30% less common - like trying, and failing, to break a habit with sheer willpower.

They'll probably change the training pipeline at some point in the future, to banish this habit, but honestly? It's such a low priority that they might take a while getting to it.

> runs as a 1 kHz leaky-integrate-and-fire (LIF) simulation

That hyphenated thing is always such a give. Clearly a thing discussed in the Claude conversation, made its way into the plan, then the readme. It always creates hyphenated phrases out of things it feels are core requirements.

I'm actually over caring about Claude's obsession with the concepts of "real", "honest", and "rather than". What annoys me the most now is how every single sentence, bullet, etc that even makes its way into the products it builds are advertisements for itself. Product settings pages now read like ads, or like something written by a student begging for extra credit from the teacher. You can't unsee it!

To be fair, as a term it has always been written as "Integrate-and-Fire" neuron model. In this case correct would be "Leaky Integrate-and-Fire" but that is nitpicking.

Resources that outdate ChatGPT: - W. Gerstner "Neuronal Dynamics" (2014): https://neuronaldynamics.epfl.ch/online/Ch1.S3.html - E. Izhikevich "Which model to use for cortical spiking neurons?" (2004): https://ieeexplore.ieee.org/abstract/document/1333071

"human written" is going away

> live *spiking* simulation

Honestly? I don't even know how to write anymore. Generative text is unequivocally nice, polished and correct. And it so happens that I do write in the same style as well. I never use LLMs to write human content, but here we are: I cannot use "honestly" anymore. And the emdash. Thank God we got "delve" back, though.

Reply on news.ycombinator.com