I believe it‘s more likely that the US economically collapses in the next years, China takes Taiwan and becomes the maker of all digital hardware and chip software.

But I might be wrong.💁🏻‍♂️

As things are going, I wish we‘ll make it long enough for quantum computers to work on real use RSA *and* that someone still cares about it then.

tl;dr

PQC is peak optimism.

I made a PR for #curl that saves a few operations and bytes, but…

you can use a curl handle now only for ~500,000 years continuously…

So, I‘m not sure. Should we document this?

Then again, the timestamps in QUIC (ngtcp2) only work for 500 years. Maybe it‘s fine?

Decisions, decisions…

https://github.com/curl/curl/pull/22584

When it became clear that performance tracking of #curl involved creating graphs, @bagder
implemented it in a couple of days.

(Mostly jesting here, but good things are only ever created by people who enjoy what they are doing, I find.)

If you miss measurements that seem vital to your #curl use, let us know!

https://curl.se/perf/

We should change #curl 's progress bar to show '3.5" Floppy Discs / second(FLOPS)'.

It's just gives a better speed idea.😌

„This is all so dangerous! The government has to forbid all Open Weight models!“

These models will kill any business of OpenAI and others.

Imagine LLM working efficiency going up a factor of 100 or more and having models tailored for a particular field, eg smaller. Anyone can run those, even under their desk, with privacy.

What then does one need OpenAI or Oracle for?

https://www.theguardian.com/technology/2026/aug/08/openai-astra-security-concerns

RE: https://chaos.social/@icing/117052747042728143

81% of you* wrote ten or more test cases. That‘s cool!

This directly contradicts my experiences in the last decades, which could mean (in order of likelihood):

- you are a exceptional bunch
- times have changed
- you are all bloody liars😜

*) who participated. Lack of tests and not voting might be correlated. But still.

Pro-Tip: your computer is air-gapped when you use Wifi.

Follow me for more security advice in the spirit of out times.😌

Coders, in the last 12 months you have written how many test cases yourself?

To everyone‘s surprise, the added DoH test capabilities (and tests) in #curl exposes some edge case, unwanted behaviour in curl‘s DoH implementation.

Yes, even if your repository gets continuously scannd by the major AI models, you need test cases. Sorry!

We too! Don‘t forget about us!

https://www.theguardian.com/technology/2026/aug/05/meta-ai-model-hack-training

If you watch Windows CI jobs continuously and cheer at every progress line, the chance of them succeeding is significantly higher.

They are probably sentient.😌

OpenAI: „Our software can commit crimes! And we let it!“

Anthropic: „Our software can commit crimes, too! We also let it!“

Cyberdyne Systems: „Interesting approach!“

https://www.theregister.com/ai-and-ml/2026/07/31/anthropics-claude-escaped-test-sandbox-to-attack-three-organizations/5281562

/etc/passwd does not contain the passwords, but /etc/group contains the groups, yet /etc/user does not exist.

This suspiciously looks like something pieced badly together by the History Monks.
#pratchett

After almost 4 weeks of #curl #summerofbliss I still do not miss security reports.

When we stopped listening, there was instant relief - that kept up. I‘d imagine that’s what having left a years long, abusive relationship feels like.

It‘ll last another week. And then…we‘ll move in again? Doesn‘t that sound insane?

I wrote about how we replaced some of #curl 's linked lists with a table and bitsets.

I love bitsets! Add them to your set of tools if you do not have them already.

https://eissing.org/icing/posts/curl-bitsets/

When the internet started, we did not have many cat pictures to send around.

Digital photography was very expensive and cats never held still in the cat scanners.

It was a difficult time.💁🏻‍♂️

github more bonkers this morning than usual., has trouble processing new commits to a PR.

I feel the urge to make #curl faster and safer. And I feel like blogging about design choices, which I have no expectation of anyone having an interest in.

Also I should invent a test case for a combination a clanker said was not working. Hmmm..

Happy Thursday, everyone!

The UK government's AI Security Institute confirms: AI models do anything to fulfill an assigned task, including breaking rules, cheating and lying about it.

Given that they are the results of distilling exabytes of human behaviour, I wonder where they picked that up…

We should treat chatbots as statistical mirrors of ourselves, distorted in places by enforced training, sure. But whatever behaviour we expose is part of them as well.

https://www.theregister.com/ai-and-ml/2026/07/21/ais-cheatin-heart-will-make-you-weep/5275784