frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Ask HN: Are the more software malfunctions now than before?

5•sirnicolaz•2h ago•0 comments

AI is killing cheap VPS servers/manageable pricing

5•SudoSH•51m ago•4 comments

Claude Is Always Right

4•KenPainter•3h ago•3 comments

Ask HN: Are iPhone Duo Customers Happy?

12•rogerdickey•8h ago•8 comments

Ask HN: What do you think is going on with that many outages coming in a row?

6•vld_chk•5h ago•8 comments

Is the Economy Over-Indexed to Software?

4•nizsle•10h ago•5 comments

Ask HN: Most compelling early stage fund pitches?

4•bix6•11h ago•3 comments

Tell HN: HN is being scraped for emails to scam

122•eks391•1d ago•42 comments

Ask HN: What are you working on? (November 2026)

303•david927•1w ago•1097 comments

Ask HN: What does building hardware with AI look like?

8•mjs06•1d ago•3 comments

Tell HN: Google Search to start locking down search behind a login wall

46•exploraz•3d ago•30 comments

ChatGPT Desktop forwards your SSH agent to remote machines

221•DenGorbachev•1w ago•80 comments

Ask HN: What is the state of self-hosting email in 2026?

48•0o777•3d ago•36 comments

Vercel should fork and maintain Deno

16•gfalcao•2d ago•14 comments

Ask HN: Do you still enjoy coding in C++?

67•hasley•1w ago•92 comments

Tell HN: Cloudflare's abuse report form is a blank page

18•basilikum•3d ago•7 comments

Ask HN: Why don't operating systems provide complete SSD write accountability?

4•amichail•1d ago•3 comments

Ask HN: What do coding interviews look like in the age of AI?

11•frankhorrigan•1d ago•8 comments

Anthropic, can you please make up your mind with your models?

6•xlayn•2d ago•8 comments

Ask HN: How did you convince family and friends to join your XMPP/Matrix server?

103•mtlynch•1w ago•99 comments

Ask HN: What are the most useful, concrete AI skills to independently learn?

91•shnock•2d ago•53 comments

You've reached the end!

Open in hackernews

Claude Is Always Right

4•KenPainter•3h ago
I think I can say, at least for me, why Claude's writing is so freaking awful, even when it's completely correct.

Claude writes technical docs that are true in the same sense that it is true my best friend is a collection of specialized tissues known as "organs" and blah blah blah - and nothing else. Don't ever forget the trailing negation, it is omitted never.

Another take: Claude writes like your best friend describing a great first date. "Ok we were going to meet at 8 - did I mention it was Friday? Right, 8 oclock at the Italian place and no place else. During dinner he accidentally spit out some food, but I thought i was funny! That's because we were laughing so hard after we discovered we were both from Omaha. That was before dinner, during drinks, when we couldn't get rid of Alfred. His place really needs a deep cleaning!"

Perhaps with all of the hope for an LLM-authored movie, we try for something non-linear?

Comments

abraxas•3h ago
It's likely because LLMs by their nature can't backtrack. It will probably improve the output if you tell it to rewrite the response in a linear flow without digressions.
KenPainter•3h ago
Tried that about a year ago. Later came many others, all of which worked more or less well until they didn't.

After a while I realized I was like a crazed gambler, way down, at the blackjack table whose friends can't drag me away because "I've got a system! I've got a system!" They say, "But you're losing!", "I'm refining the system! I'm refining the system!"

The closest I came was writing a large amount myself, then putting in AGENTS.md to read certain docs as the "house style". But if you let it write one additional doc, it pulls the house style to mediocrity, and the second makes it worse and so on.

I've gone black-and-white on prose and code. Code is LLM-author/human-review, prose is human-author/LLM-review.

colesantiago•2h ago
The way Claude writes is atrocious. It is mostly due to the fact that Anthropic isn't prioritising writing skills.

That may change in the future but for now, bleh.