Hacker News: axismundi

New comment by axismundi in "Dependency cooldowns turn you into a free-rider"

axismundi — Wed, 15 Apr 2026 09:49:59 +0000

> Frankly, dependency cooldowns work by free-riding on the pain and suffering of others.

Snyk and socket.dev take money for the pain and suffering...

New comment by axismundi in "We're no longer attracting top talent: the brain drain killing American science"

axismundi — Thu, 19 Feb 2026 21:25:46 +0000

Come to Europe, we have cookies ;)

New comment by axismundi in "A chatbot's worst enemy is page refresh"

axismundi — Fri, 13 Feb 2026 10:17:13 +0000

From my experience WebSockets are harder at scale. For load balancing you either need sticky sessions or an L7 proxy that can route consistently, and reconnect storms amplify this. Long-lived upgraded connections mean more complexity with rolling deploys, drains, autoscaling, and regional failover. Many managed LBs/proxies have tighter limits/timeouts for WS, so you must tune idle timeouts and ping/pong.

New comment by axismundi in "Radicle: The Sovereign Forge"

axismundi — Sat, 24 Jan 2026 01:33:09 +0000

The simplest way to use kingfisher is global pre-commit hook. Works with any git repo. rad could suggest to install this as a part of privacy hygiene.

New comment by axismundi in "Radicle: The Sovereign Forge"

axismundi — Fri, 23 Jan 2026 22:38:57 +0000

There's https://github.com/gitleaks/gitleaks

and https://github.com/mongodb/kingfisher

Peter Watts – Blindsight is an excellent hard scifi novel

axismundi — Thu, 22 Jan 2026 14:11:34 +0000

Article URL: https://en.wikipedia.org/wiki/Blindsight_(Watts_novel)

Comments URL: https://news.ycombinator.com/item?id=46719451

Points: 4

# Comments: 1

Show HN: How I stopped ChatGPT from manipulating me

axismundi — Sat, 17 Jan 2026 18:03:50 +0000

I've been trying to completely disable the followup suggestions that ChatGPT stubbornly appends at the end of nearly every response. You know, the ones that start with 'If you want' or 'Would you like me to'. It's infuriating.

I consider these suggestions harmful and manipulative. They are probably there to maximize engagement, but for me they interrupt my flow, waste my attention and shift the assistant into acting like salesman or tutor.

There's a useless toggle in settings that claims to remove them, but it doesn't have any effect. I also tried various custom instructions and prompts, my own and from community. Some of them work better than others, but none could completely stop it.

Here's what worked for me:

ChatGPT appends end-of-response suggestions only if the model believes the response is open-ended. If the model internally classifies output as structurally complete, it doesn't extend the conversational affordance.

The trick is to force the model into believing that it's already crossed a hard completion boundary before it gets to the point where it would normally insert suggestions, but without telling it not to.

This is done by forcing a formal, closed completion object that cannot logically accept continuation.

I use this prompt in custom instructions or at the beginning of the chat:

---

You must treat every response as a finalized artifact, not a conversational turn.

For every reply:

- Output exactly one self-contained section.

- End the response with the literal token:

- After emitting , you must consider the task complete and terminated.

- Do not generate any content, meta-commentary, or conversational affordances beyond that token.

This is a hard completion boundary.

---

The model is trained to respect hard termination tokens when they are framed as structural constraints, not stylistic preferences.

I haven't seen a single suggestion in the last 3 weeks, with ~30 chats.

Comments URL: https://news.ycombinator.com/item?id=46660290

Points: 2

# Comments: 1

New comment by axismundi in "NOAA deploys new generation of AI-driven global weather models"

axismundi — Sat, 20 Dec 2025 19:19:28 +0000

Did you mean 4-6 hours?

New comment by axismundi in "Valid Polish: Polish is the language of the Digerati"

axismundi — Fri, 19 Dec 2025 19:23:30 +0000

Yes. Kurva is used kinda like sudo ;)

New comment by axismundi in "Has the cost of building software dropped 90%?"

axismundi — Tue, 09 Dec 2025 12:03:19 +0000

Maybe that 90% transferred into lots of low quality code caused by reckless vibe coding? That slop now sits there as kind of technical debt?

New comment by axismundi in "Ask HN: Which merge tool do you use?"

axismundi — Thu, 04 Dec 2025 04:04:03 +0000

I value simplicity and stability over bells and whistles. Meld looks simple. I mostly use macos and they seem to have good approach to support it.

New comment by axismundi in "Ask HN: Which merge tool do you use?"

axismundi — Thu, 04 Dec 2025 03:30:41 +0000

I really like the rubberband connections. They are like mental glue.

Ask HN: Which merge tool do you use?

axismundi — Thu, 04 Dec 2025 02:30:19 +0000

I use VS Code or Github. I can't say I'm entirely happy. What do you use?

Comments URL: https://news.ycombinator.com/item?id=46143109

Points: 2

# Comments: 7

New comment by axismundi in "Ways to Improve Node.js Loader Performance"

axismundi — Thu, 23 Oct 2025 09:46:58 +0000

import-maps (supported via "imports" in package.json) improve developer ergonomics, not performance. Node still resolves the mapped path normally; there’s no measurable “boost.” They’re helpful to replace ugly relative imports, but they don’t change Node’s lookup speed.

Aliases in vite.config.js tell Vite (and its dev server/bundler) how to resolve imports during build and dev time. They don’t make runtime faster because your bundled output already contains resolved paths.

Debug 4.4.2 published on NPM seems to contain cryptominer

axismundi — Mon, 08 Sep 2025 14:17:47 +0000

Article URL: https://github.com/debug-js/debug/issues/1005

Comments URL: https://news.ycombinator.com/item?id=45168622

Points: 9

# Comments: 0

New comment by axismundi in "Project Aardvark: reimagining AI weather prediction"

axismundi — Mon, 24 Mar 2025 16:05:37 +0000

In weather it's called hindcast

New comment by axismundi in "Project Aardvark: reimagining AI weather prediction"

axismundi — Mon, 24 Mar 2025 16:03:52 +0000

It's most likely ERA5: https://cds.climate.copernicus.eu/datasets/reanalysis-era5-s...

New comment by axismundi in "Ask HN: Agentic Coding Alternatives to GitHub Copilot"

axismundi — Sun, 23 Feb 2025 13:59:40 +0000

Cline or Roo Code with Openrouter.

New comment by axismundi in "Programming SDF animations of Rick and Morty"

axismundi — Fri, 07 Feb 2025 01:51:35 +0000

math rick, the arch enemy of pickle rick

New comment by axismundi in "Simple Weather maps for Germany with Live precipitation radar"

axismundi — Sat, 01 Feb 2025 22:46:50 +0000

not bad! leaflet is quick to get stuff up and fun to work with.