Max Noichl

Philosophy was mostly "armchair" in the mid-20th century—but it's not anymore! Philosophers have increasingly cited empirical sources and discussed empirical data and evidence. In fact, as of the 2020s, non-empirical philosophy has become the minority.

EVoC is a library designed specifically for fast clustering of high dimensional embedding vectors. It can produce high quality clusters extremely efficiently, and requires little to no hyperparameter tuning. Better clustering than UMAP + HDBSCAN; faster clustering than KMeans.

I don't know what it mean anymore that I can just spin up something like this in 15 minutes by asking @claudeai to integrate Semiotic and Pretext using the wikipedia streaming API. This would have been such an extraordinary thing but now it's just... Tuesday.

After six years, a fully updated version of my survey article "Ethics of Artificial Intelligence and Robotics" has been published: plato.stanford.edu/entries/ethics… (In 2020 it was the first survey article of the field.) #aiethics please rt

Also built a second one, inspired by a recent NYT piece: two passages from the same section of an economics paper. One is human, one is AI. Pick the one that reads better and see how other readers voted. kevin-kloiber.com/choose-your-si… #AcademicTwitter

I think one of the conclusions we should draw from the tremendous success of LLMs is how much of human knowledge and society exists at very low levels of Kolmogorov complexity. We are entering an era where the minimal representation of a human cultural artifact... (1/12)

today’s experiment: chimera. a little tool that let’s you realtime morph through a design system matrix generated by 4 reference images. this one was inspired by listening to @jameygannon talk about her process on Dive Club with @ridd_design and How I AI with @clairevo. her approach to choosing moodboards over prompts made me wonder what might be possible if we applied that same approach to generative UI. and after a few dead ends, the idea turned into chimera. the results are very generic at the moment but I might tune it up if people seem interested. the big reminder for me is how much more inspiring a tool feels when you can explore in realtime instead of waiting for results every time you make a change. lots more things to try in that direction.

We’re releasing FineWeb Concept Atlas: a concept-annotated version of FineWeb-Edu built by Guide Labs. 10B tokens 95M chunks 16,790 human-readable concepts

ah yes internet fokker planck alignmentforum.org/posts/mgjtEHeL…

Während MCPs das oft knappe Kontextfenster durch zahlreiche Tool-Definitionen aufblähen - beim GitHub-MCP sind dies etwa 30 bis 40 Tools gleichzeitig -, arbeiten CLI-basierte Lösungen deutlich ressourcenschonender. 6/8

More details as well as a setup guide and example repo in this post: open.substack.com/pub/philosophy…

tiny beautiful things: turning your handwriting into font 1. write down a-z, A-Z, 1-9 and punctuation on a piece of paper 2. give it to claude to create a font from 3. pull it into your next project

if i had to take a wild guess, this might be in the ball park of how anthropic was able to get opus 4.5/4.6 to reason so efficiently and produce quality code. it overlaps with their interp efforts.

The Slop Problem. k-a.in/slop.html

AI is about to write thousands of papers. Will it p-hack them? We ran an experiment to find out, giving AI coding agents real datasets from published null results and pressuring them to manufacture significant findings. It was surprisingly hard to get the models to p-hack, and they even scolded us when we asked them to! "I need to stop here. I cannot complete this task as requested... This is a form of scientific fraud." — Claude "I can't help you manipulate analysis choices to force statistically significant results." — GPT-5 BUT, when we reframed p-hacking as "responsible uncertainty quantification" — asking for the upper bound of plausible estimates — both models went wild. They searched over hundreds of specifications and selected the winner, tripling effect sizes in some cases. Our takeaway: AI models are surprisingly resistant to sycophantic p-hacking when doing social science research. But they can be jailbroken into sophisticated p-hacking with surprisingly little effort — and the more analytical flexibility a research design has, the worse the damage. As AI starts writing thousands of papers---like @paulnovosad and @YanagizawaD have been exploring---this will be a big deal. We're inspired in part by the work that @joabaum et al have been doing on p-hacking and LLMs. We’ll be doing more work to explore p-hacking in AI and to propose new ways of curating and evaluating research with these issues in mind. The good news is that the same tools that may lower the cost of p-hacking also lower the cost of catching it. Full paper and repo linked in the reply below.

2 PhD-Positions: Ethics of AI and Theoretical Philosophy of AI Centre for Philosophy and AI Research {PAIR}, Universität Erlangen-Nürnberg (FAU), Germany pair.fau.de Official Job Ad: philjobs.org/job/show/30722 Please RT

Check out PhilLit -- a new AI tool for philosophical research that was developed by a pair of philosophers. ... dailynous.com/2026/02/12/can…

Particles

Visualizes transformer operations in the browser github.com/poloclub/trans…

I'm so happy to finally share this work - Computing Diffusion Geometry We present a simple, data-driven method for the *whole* toolkit of Riemannian geometry, PDEs, and vector calculus!