Philipp Hugenroth

A hill I'll die on: Current LLM chat interfaces are a regression from GUIs. Actions that used to be links, buttons, or keyboard shortcuts are now things I have to spell out in conversation. Why?

@badlogicgames @mitsuhiko an opinion piece on os community management in the age of clakers would be interesting. you do some things different with e.g. the os vacations (or in general how to think about outside contributions).

@mitsuhiko Conversely, I feel kinda bad for not having written a blog post in 3 months. But besides some orchestration experiments, there just doesn't seem to be anything worthwhile to publish atm. Neither work I did myself, not about work others did.

I'm trying to catch up with some recent agentic engineering developments and I keep being reminded by this tweet. Barely anything these days is just people sharing ideas, people share products and vibeslopped websites. So much effort goes into marketing. x.com/mitsuhiko/stat…

fun and easy to play around with. love the halftone dots paired with the riso style. was ready to print after 15 minutes of experimenting. could directly export and print without another round of photoshop. printed at the ppppress in the mit media lab.

from @paper to paper. using halftone CMYK shader to create RISO prints.

@badlogicgames Agree. It even helps techies. There is a reason we learn frameworks and formal methods in education. A lot of us just internalised those things so much that through experience they turn into intuition/taste.

recommended reading. one of the issues of agent-based apps is that many non-techies have trouble understanding what they can actually do with it. an agent-guided "tutorial" of sorts really helps non-techies explore capabilities. those tutorials are kinda hard to design tho, because things aren't alway as deterministic as a nice learning path would require.

@xenovacom @karpathy what a week of putting things out there for web ml - GPT-OSS, RF-DETR, QED-nano, Tiny Aya - now this. respect and cool to see!

Build, train, and run LLMs — directly in your browser. Inspired by @karpathy's microgpt, I built an educational neural network builder that breaks down "mysterious" LLMs into their primitive components. Learn how LLMs are built from the ground up. Try it out yourself!

rtp: Vera Molnár, HTML Canvas: with(c.getContext("2d"))for(lineWidth=30,i=11; i--; stroke())(beginPath(),(strokeStyle=i%2?"#000":"#175DA8"),moveTo((X=85*(i+1)),0),lineTo(X,(Y=550*Math.random())-50),arc(X-15,Y,50,-Math.PI/2,Math.PI/2),moveTo(X,Y+50),ineTo(X,600));

@steipete long live the web.

apps are dead

semantic web is so Y2K

stuck in a vera molnar rabbit hole youtube.com/watch?v=HkA0hg…

Much like the switch in 2025 from language models to reasoning models, we think 2026 will be all about the switch to Recursive Language Models (RLMs). It turns out that models can be far more powerful if you allow them to treat *their own prompts* as an object in an external environment, which they understand and manipulate by writing code that invokes LLMs! Our full paper on RLMs is now available—with much more expansive experiments compared to our initial blogpost from October 2025! arxiv.org/pdf/2512.24601

I don't know why this week became the tipping point, but nearly every software engineer I've talked to is experiencing some degree of mental health crisis.

@thorstenball this. everything is different, while everything is the same.

Trying to catch my wife up on the latest in AI whenever we meet in the kitchen always ends with wife: yeah, yeah, I get it me: but *do you*?

@openrouter @arcee_ai non-paywall version: #selection-4057.431-4061.206" target="_blank" rel="nofollow noopener">archive.is/leepa#selectio…

@openrouter @arcee_ai according to forbes they are trying to collect 200m$ at 1B$ right now: forbes.com/sites/annatong…

free model provider on @openrouter as a growth hack. only yesterday @arcee_ai (only ~30m funding, ~50 people) got 50B+ prompts. not only tons of usage data, but also could help with extrapolating statistics for fundraising.

the internet is dead, long live the web.

seems to explain most of it github.com/black-forest-l…

interested in how they interpret the prompt & decide which parameters should be changeable. my naive prompt was just the first paragraph of "The Metamorphosis". directly prompting the model without the interpretation (e.g. through krea) looks vastly different.

the @bfl_ml flux playground ux is really interesting. great concept to develop intuition for how to prompt the model and visually explore specific/revertible modifications.