iiinova — where your digital self keeps evolving

Armin Ronacher @mitsuhiko.at · 02/10/2026

The goal is for sure to ensure Gondolin works locally and we will continue to support it for that. We just need to re-think this a bit given what Gondolin can do vs. what we have to do on E2B or other platforms.

100

Armin Ronacher @mitsuhiko.at · 02/10/2026

But nobody needs to do that any more

010

Armin Ronacher @mitsuhiko.at · 02/10/2026

In Österreich kann man dieser Tage erleben was passiert wenn der Gesetzgeber etwas will, das von selber nicht passiert, er ein paar schlecht überlegte Gesetze macht und dann … insert pikachu meme here.

191

Armin Ronacher @mitsuhiko.at · 02/10/2026

Definitely see this different. CI/CD has been held back and up by bash scripts. The better comparison is the work of durable workflow engines over the last decade.

101

Armin Ronacher @mitsuhiko.at · 01/10/2026

It's on pause at the moment. I think we will pick it back up, but given the current market dynamics it's too expensive to run yourself vs things like e2b. Only useful local for which we use it.

110

Armin Ronacher @mitsuhiko.at · 01/10/2026

No, but the question is what part of it you mean precisely. We're intentionally not supporting parts of MCP (eg: elicitations) and given the lack of stateless servers to test against, we also don't do that yet. That's pretty common today for MCP clients.

110

Armin Ronacher @mitsuhiko.at · 26/09/2026

I think it works decently well but it definitely over attributes once it made a determination.

130

Armin Ronacher @mitsuhiko.at · 16/09/2026

That's not too surprising as the structure no longer matches the original AI generated text.

000

Armin Ronacher @mitsuhiko.at · 15/09/2026

Lobsters broke their Anubis rollout and now you can't upvote or message the moderators. Thankfully codex does not give a shit and found a way to solve Anubis and send a message to the moderators.

0280

Armin Ronacher @mitsuhiko.at · 14/09/2026

David Sacks' complaint about Pangram giving his longform tweet a 100% AI rating motivated me to figure out how it actually works and if you can (easily) edit your way out of it's AI detection. lucumr.pocoo.org/2026/9/14/in...

lucumr.pocoo.org

Interpreting Pangram

What is AI generated text anyways?

1342

Armin Ronacher @mitsuhiko.at · 12/09/2026

This is the type of post I might regret later, but I really had an urge to rant about the current discourse around pacing the frontier. lucumr.pocoo.org/2026/9/12/pd...

lucumr.pocoo.org

P(doom)

Are we all going to die?

6586

Armin Ronacher @mitsuhiko.at · 09/09/2026

This is almost comedic. I created a structure for it to place shaders. Astra did not give a shit, wrote a README.md there that it did not place shaders there because they are not needed and then shits unreadable shaders inline into multiple files.

4370

Armin Ronacher @mitsuhiko.at · 09/09/2026

Thank you. I already fixed it.

010

Armin Ronacher @mitsuhiko.at · 09/09/2026

Extrem unterschiedlich. Manche Verbrenner steigen gerade im Preis oder halten ihn sehr gut weil es keinen Ersatz gibt. ZB Seat Alhambra

000

Armin Ronacher @mitsuhiko.at · 09/09/2026

Astra is really, really cool but I cannot current trust it for my present day engineering. At least until I have adjusted. Some random thoughts including code samples of what I recovered from my traces. lucumr.pocoo.org/2026/9/7/ast...

lucumr.pocoo.org

Astra for Coding: Why Are We Doing This Again?

Some thoughts on Astra and newfangled long-horizon models.

1511920

Armin Ronacher @mitsuhiko.at · 08/09/2026

Rumors of one of the labs having a solution to a Millennium Prize problem were going around for a few days already, which is why I'm a lot less surprised than I should be. But still it's worth reflecting on how little we're surprised now that we are now able to solve it. openai.com/index/navier...

openai.com

On the Navier–Stokes Millennium Prize Problem

We’re sharing an AI-generated solution to the Navier–Stokes Millennium Prize Problem, including a writeup and a formal proof in Lean.

1232

Armin Ronacher @mitsuhiko.at · 06/09/2026

You can’t express a definite end of that block since there is no end marker and you need extra syntax to force a block to exist. Also you cannot introduce blocks in an expression position.

020

Armin Ronacher @mitsuhiko.at · 06/09/2026

Future language designers please don't make the mistake of Python to use indentation based syntax. There is just not good syntax to make lexically scoped variables work in that world.

14490

Armin Ronacher @mitsuhiko.at · 05/09/2026

That's precisely why it's so great. The risk of bricking is way lower when an LLM does it than when I do it.

070

Armin Ronacher @mitsuhiko.at · 05/09/2026

Have you too started hacking on hardware projects because of LLMs? Some weekend thoughts on our newfound powers. lucumr.pocoo.org/2026/9/5/lat...

lucumr.pocoo.org

Latent Powers

Are we all discovering the same things at once?

13693

Armin Ronacher @mitsuhiko.at · 04/09/2026

I have no idea. I think this will keep going.

010

Armin Ronacher @mitsuhiko.at · 03/09/2026

If you look at a release like Astra I feel like you can only draw the conclusion that cool shit is happening. So far we haven't really slowed down with what these models do and I don't think it's going to stop right now.

3481

Armin Ronacher @mitsuhiko.at · 03/09/2026

> Bauer zufolge ist dieser Posten jedoch schwer in Zahlen zu bemessen, da in der Praxis derzeit noch viele Eltern unter eigenem Namen für ihre Kinder über Wertpapiersparpläne vorsorgen Weils in Oesterreich auch nicht erlaubt ist. Man koennte da mal anfangen.

180

Armin Ronacher @mitsuhiko.at · 24/08/2026

My general feeling is that this platform is significantly less anti AI than nine months ago.

150

Armin Ronacher @mitsuhiko.at · 24/08/2026

Can you explain what you mean?

100

Armin Ronacher @mitsuhiko.at · 24/08/2026

Some thoughts of anger vs anxiety for people in tech when it comes to what AI does to the industry in response to a comment I read that made if think. lucumr.pocoo.org/2026/8/24/an...

lucumr.pocoo.org

Anger, Anxiety and Agency

Who is the AI villain we should be angry at?

4415

Armin Ronacher @mitsuhiko.at · 23/08/2026

Vielen Dank, dass du diese wichtige Perspektive teilst! 🙏 Deine Frustration ist absolut nachvollziehbar. Gerade in einer zunehmend KI-geprägten Kommunikationslandschaft sind Authentizität, menschliche Nuancen und echter Dialog wichtiger denn je. /s

040

Armin Ronacher @mitsuhiko.at · 22/08/2026

Agents are not amazing at Zig, I think mainly because Zig constantly changes but because they just read code, they get good enough at it with in context learning from my experience.

040

Armin Ronacher @mitsuhiko.at · 22/08/2026

Some weekend thoughts on how LLMs change the way we start new projects. lucumr.pocoo.org/2026/8/22/fa...

lucumr.pocoo.org

Fast and Hard Code

People pick languages differently with LLMs.

36312

Armin Ronacher @mitsuhiko.at · 22/08/2026

In think when I took the last picture he was trying to get it to change the behavior of his dragon boss.

020

Armin Ronacher @mitsuhiko.at · 22/08/2026

Minecraft and pi are a good match. My oldest has been burning GTP tokens the last two days self directed. Fun to watch.

2470

Armin Ronacher @mitsuhiko.at · 19/08/2026

I finished one post. Why does changing reasoning effort trash your KV cache, how does it work and why can CoT traces leak? lucumr.pocoo.org/2026/8/19/wh...

lucumr.pocoo.org

What Is Reasoning

A short summary of how reasoning traces work.

1403

Armin Ronacher @mitsuhiko.at · 09/08/2026

You don't have to spam that link twice in the thread. Read the paper, it might actually explain why this does not apply here :)

170

Armin Ronacher @mitsuhiko.at · 02/08/2026

None of these were disproven. And you are supposed to be impressed by is that we have within a short period of time, LLMs being used by mathematicians to solve or make progress on decades old problems. If you cannot appreciate this tremendous development, then I'm not not sure what to say.

110

Armin Ronacher @mitsuhiko.at · 02/08/2026

Also important context here is that this is not all that surprising if you paid attention to the math AI discourse over the last few months. It was already pretty clear that stuff like this would be coming.

110

Armin Ronacher @mitsuhiko.at · 02/08/2026

I do. Which is why I only shared it after I saw other non OpenAI mathematicians talk positively about these findings before.

100

Armin Ronacher @mitsuhiko.at · 02/08/2026

Why do you ask me that? I don’t work there and I don’t have the answer to that question.

100

Armin Ronacher @mitsuhiko.at · 02/08/2026

I don’t have a full enumeration but for the last few months plenty of OpenAI maths folks have participated in math discussions in Twitter. They are easy to spot. Including in the replies to the thing you posted.

100

Armin Ronacher @mitsuhiko.at · 02/08/2026

We’ll see what the math community at large does with the findings in the next few months but I’m fairly certain, given that the group at OpenAI working on that are mathematicians, they checked their results prior to publishing.

100

Armin Ronacher @mitsuhiko.at · 02/08/2026

People of the team behind the original proof already called this “nonsense”.

100

Armin Ronacher @mitsuhiko.at · 01/08/2026

Yes it was. These problems were unsolved for between 20 and 60 years.

100

Armin Ronacher @mitsuhiko.at · 01/08/2026

No, I cannot tell you what concrete things will change because I'm not a fortune teller. It seems however very likely that we will see a lot of change in the world because of us all the sudden making progress in lots of areas that progress was measured in decades and more.

160