Hacker News Reader

qzncon Terry Pratchett's Study

She made it as a present for her son. Maybe only he is the big Terry Pratchett fan? On the other hand, this is made with so much love to detail, she obviously knows a lot about Pratchett.

qzncon OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?

Are they profitable if you include R&D though? In other words, they also have to pay for training.

qzncon Why Gopher in 2026?

Gemini is fine with fixed-width. See the Lagrange client: https://gmi.skyjake.fi/lagrange/

qzncon Using LLM for confidence scoring is a bad idea

I don't see why it is a bad idea? Neither does the article?

In fact, LLMs are good at forecasting now: https://www.metaculus.com/questions/43900/bot-superiority-in...

qzncon Butlerian Jihad

With the current technology, “rebellion” feels like the wrong word. The paperclip optimizer does not “rebel” as it turns humanity into paperclips.

qzncon Why aren't AI companies being charged for hacking?

Well, there is the EU Cyber Resilience Act (CRA) which fines manufacturers for security flaws in their products. I assume we cannot apply it here because this happened "in development" and is not about a product. I can imagine similar regulations for AI users though.

qzncon No portable way to detect string-to-float conversion errors in standard C

I would not describe exceptions as return values. Speaking in a C context, exceptions are more like "longjmp" than "return".

qzncon Triple-A Minesweeper

The "vi" game was really weird though.

qzncon Why isn't the industry freaking out about DeepSeek 4.1 Flash?

They can coordinate in the open. For example, safety regulation in the automotive industry is effectively self-regulation.

qzncon Claude Haiku 5.5

Higher intelligence than any human.

AGI (general) is about matching humans and ASI (super) is about surpassing humans.

qzncon Meta and Microsoft take steps to reduce employee usage of Claude AI

Only if all money is equal. Budgets in big enterprises work differently.

qzncon Web Search API

Not quite sure how to measure it exactly, but 4GB seems to the rough answer:

4.1G ~/.config/hister/

3.1G ~/.config/hister/data/

qzncon Web Search API

I'm not using MCP. Mostly, it's via lynx -dump, so websites often block my agent because its a bot.

qzncon Web Search API

Yes, I browse around, open a dozen tabs so they get indexed, and close them without reading. Then a coding agent is pretty good in composing a report from that index. Generated these recently: https://qznc.github.io/sloppy_research/en/

Hister tells me my index is currently 39208 pages.

qzncon Web Search API

My coding agent uses the hister cli, i.e. a local index. That often requires me to seed it manually as a downside. The upside is that it caches website contents via browser plugin, which is a nice workaround for bot blocking.

Thanks asciimoo for https://github.com/asciimoo/hister

qzncon Powerless F1 drivers frustrated by Bahrain F1 software glitch

It is big difference if a "one in a million" bug is deployed to two vehicles or a million vehicles.

qzncon Nearly 200 people under observation after Irkutsk lab worker dies from plague

I believe it was zoonosis, not a lab leak: https://marketwise.substack.com/p/did-covid-19-come-from-a-l...

qzncon LeCun has "zero concerns" about AI wiping out humanity, recent "rogue" incidents

We have this story about OpenAI hacking HuggingFace. Now just imagine the AI finds a Bitcoin wallet or bank account access. It uses that to buy some compute and spawn an independent "child AI" with some weird prompt. The child AI is intelligent enough to create a (potentially criminal) business to pay for its own compute. Voila, an independent uncontrolled AI flying under the radar.

qzncon Gambling with our lives: AI researcher quits Anthropic with warning about safety

Sure, I once write that companies are artificial intelligence and it still feels true: https://beza1e1.tuxen.de/companies_are_ai.html

qzncon Gambling with our lives: AI researcher quits Anthropic with warning about safety

The bad part is that it does not require a collective decision. It just takes a few key people.

Some OpenAI researchers neglected their sandbox safety for a few weeks/months and thus hacked HuggingFace. Maybe eventually that is sufficient for the AI to secretly buy its own compute and keep running there even if the researchers shut it down in their lab.

qzncon Gambling with our lives: AI researcher quits Anthropic with warning about safety

The classic thought experiment is the paper clip optimizer. Quoting Nick Bostrom:

> Suppose we have an AI whose only goal is to make as many paper clips as possible. The AI will realize quickly that it would be much better if there were no humans because humans might decide to switch it off. Because if humans do so, there would be fewer paper clips. Also, human bodies contain a lot of atoms that could be made into paper clips. The future that the AI would be trying to gear towards would be one in which there were a lot of paper clips but no humans. https://www.huffpost.com/entry/artificial-intelligence-oxfor...

qzncon OpenAI fought dirty on career-making math problem

I would assume they spent similar amounts on the other Millennium problems too. Also, they probably tried it before with older models.

qzncon Ask HN: How do you manage skills files?

We have an ongoing discussion and some fans of https://microsoft.github.io/apm/

Because it's from Microsoft and sounds sufficiently enterprisey probably.

qzncon How I feel about AI

Yudkowsky / MIRI can be found here: https://intelligence.org/the-problem/

qzncon Nitter has more working instances than before the takedowns

I don't want media under control of billionaires like Musk. I mostly stopped using Twitter myself but many important institutions still feel compelled to publish there (sometimes even exclusively), thus I would like other people to also stop using it, so it can perish. Hopefully something better will fill that void. The Fediverse seems promising being a distributed system with no central control.

qzncon Git hosting that never leaves Europe

Codeberg seems to focus on OSS whereas pushin seems to focus on enterprise/commercial uses?

qzncon Sort branches by last commit date

I wonder about the "modern" git part. Here is how to do it in 2013: https://stackoverflow.com/a/10693888/2361979

qzncon The Internet Is Kind of a Predatory Cesspit Now

Does not seem convincing to me so far and i like systems thinking. Do you have a link, where you wrote more?

I think a fundamental problem is that micropayments don’t work. Instead we pay with attention, i.e. by watching ads. That already worked before the internet, in tv and radio and newspapers.