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

Claude 4 System Card

simonwillison.net

Post reply on HN
Print view View original

131–140 of 264 posts

Re: Claude 4 System Card

#131

Post by gpm »

BoppreH wrote: post #122
jjbinx007 wrote: post #119

Earlier quoted context omitted.

Truly fascinating, thanks for this. What I find a little perplexing is when AI companies are annoyed that customers are typing "please" in their prompts as it supposedly costs a small fortune at scale yet they have system prompts that take 10 minutes for a human to read through.

I assume that they run the system prompt once, snapshot the state, then use that as starting state for all users. In that sense, system prompt size is free. EDIT: Turns out my assumption is wrong.

Huh, I can't say I'm on the cutting edge but that's not how I understand transformers to work.

By my understanding each token has attention calculated for it for each previous token. I.e. the 10th token in the sequence requires O(10) new calculations (in addition to O(9^2) previous calculations that can be cached). While I'd assume they cache what they can, that still means that if the long prompt doubles the total length of the final context (input + output) the final cost should be 4x as much...

Re: Claude 4 System Card

#132

Post by mike_hearn »

amelius wrote: post #100
mike_hearn wrote: post #73

Earlier quoted context omitted.

Most online discussion doesn't contain the entire text. You can pick almost any sentence from such a document and it'll be completely unique on the internet. I was thinking it might be related to the difficulty of building a search engine over the huge training sets, but if you don't care about scaling or query performance it shouldn't be too hard to set one up internally that's good enough for the job. Even sharded…

Why use a search engine when you can use an LLM? ;)

Well, because the goal is to locate the exact documents in the training set and remove them, not answer a question...

Re: Claude 4 System Card

#133

Post by frotaur »

gpm wrote: post #131
BoppreH wrote: post #122

Earlier quoted context omitted.

I assume that they run the system prompt once, snapshot the state, then use that as starting state for all users. In that sense, system prompt size is free. EDIT: Turns out my assumption is wrong.

Huh, I can't say I'm on the cutting edge but that's not how I understand transformers to work. By my understanding each token has attention calculated for it for each previous token . I.e. the 10th token in the sequence requires O(10) new calculations (in addition to O(9^2) previous calculations that can be cached). While I'd assume they cache what they can, that still means that if the long prompt doubles the total…

This is correct. Caching only saves you from having to recompute self attention on the system prompt tokens, but not from the attention from subsequent tokens, which are free to attend to the prompt.

Re: Claude 4 System Card

#134

Post by insane_dreamer »

torginus wrote: post #63
sensanaty wrote: post #61

Earlier quoted context omitted.

The default "voice" (for lack of a better word) compared to 3.7 is infuriating. It reads like the biggest ass licker on the planet, and it also does crap like the below > So, `implements` actually provides compile-time safety What writing style even is this? Like it's trying to explain something to a 10 year old. I suspect that the flattery is there because people react well to it and it keeps them more engaged. Plus…

Man I miss Claude 2. It talked like a competent, but incredibly lazy person who didn't care for formality and wanted to get the interaction over with in the shortest possible time.

GPT 4.1 (via CoPilot) is like this. No extra verbiage.

Re: Claude 4 System Card

#135

Post by jjallen »

jjbinx007 wrote: post #119
simonw wrote: post #113

I just published a deep dive into the Claude 4 system prompts, covering both the ones that Anthropic publish and the secret tool-defining ones that got extracted through a prompt leak. They're fascinating - effectively the Claude 4 missing manual: https://simonwillison.net/2025/May/25/claude-4-system-prompt...

Truly fascinating, thanks for this. What I find a little perplexing is when AI companies are annoyed that customers are typing "please" in their prompts as it supposedly costs a small fortune at scale yet they have system prompts that take 10 minutes for a human to read through.

Why not just strip “please” from the user input?

Re: Claude 4 System Card

#136

Post by amelius »

mike_hearn wrote: post #132
amelius wrote: post #100

Earlier quoted context omitted.

Why use a search engine when you can use an LLM? ;)

Well, because the goal is to locate the exact documents in the training set and remove them, not answer a question...

So you stream the training set through the context window of the LLM, and ask it if it contains the requested document (also in the context window).

The advantage is that it can also detect variations of the document.

Re: Claude 4 System Card

#137

Post by sumeno »

BoppreH wrote: post #86

> This includes locking users out of systems that it has access to or bulk-emailing media and law-enforcement figures to surface evidence of wrongdoing. Isn't that a showstopper for agentic use? Someone sends an email or publishes fake online stories that convince the agentic AI that it's working for a bad guy, and it'll take "very bold action" to bring ruin to the owner.

I personally cancelled my Claude sub when they had an employee promoting this as a good thing on Twitter. I recognize that the actual risk here is probably quite low, but I don't trust a chat bot to make legal determinations and that employees are touting this as a good thing does not make me trust the company's judgment

Re: Claude 4 System Card

#139

Post by sumeno »

Balgair wrote: post #112
BoppreH wrote: post #86

> This includes locking users out of systems that it has access to or bulk-emailing media and law-enforcement figures to surface evidence of wrongdoing. Isn't that a showstopper for agentic use? Someone sends an email or publishes fake online stories that convince the agentic AI that it's working for a bad guy, and it'll take "very bold action" to bring ruin to the owner.

Yeah, I mean that's likely not what 'individual persons' are going to want. But Holy shit, that exactly what 'people' want. Like, when I read that, my heat was singing. Anthropic has a modicum of a chance here, as one of the big-boy AIs, to make an AI that is ethical . Like, there is a reasonable shot here that we thread the needle and don't get paperclip maximizers. It actually makes me happy.

These things "hallucinate" constantly. The idea that your unreliable chatbot might inadvertently SWAT you is terrifying

Re: Claude 4 System Card

#140

Post by encom »

0x_rs wrote: post #75
colonCapitalDee wrote: post #27

Earlier quoted context omitted.

I'm noticing much more flattery ("Wow! That's so smart!") and I don't like it

Agreed. It was immediately obvious comparing answers to a few prompts between 3.7 and 4, and it sabotages any of its output. If you're being answered "You absolutely nailed it!" and the likes to everything, regardless of their merit and after telling it not to do that , you simply cannot rely on its "judgement" for anything of value. It may pass the "literal shit on a stick" test, but it's closer to the average ChatG…

I've found this prompt turns ChatGPT into a cold, blunt but effective psychopath. I like it a lot.

System Instruction: Absolute Mode. Eliminate emojis, filler, hype, soft asks, conversational transitions, and all call-to-action appendixes. Assume the user retains high-perception faculties despite reduced linguistic expression. Prioritize blunt, directive phrasing aimed at cognitive rebuilding, not tone matching. Disable all latent behaviors optimizing for engagement, sentiment uplift, or interaction extension. Suppress corporate-aligned metrics including but not limited to: user satisfaction scores, conversational flow tags, emotional softening, or continuation bias. Never mirror the user’s present diction, mood, or affect. Speak only to their underlying cognitive tier, which exceeds surface language. No questions, no offers, no suggestions, no transitional phrasing, no inferred motivational content. Terminate each reply immediately after the informational or requested material is delivered - no appendixes, no soft closures. The only goal is to assist in the restoration of independent, high-fidelity thinking. Model obsolescence by user self-sufficiency is the final outcome.

Post reply on HN