I was like is Nijntje researching heat kernels now??

"Do you expect authors to check every citation is real?" is an amazing sentiment exposing something very deep about research culture. Papers are meant to be written and metricized not read and used,

Hallucinating citations is cheap. What do we do about a paper misinterpreted (including making up that they described an algorithm they didn't) which gets replicated in infinite subsequent citations. I remember this prevalent phenomenon from the 2010s,

How come my (pseudo)papers never come up in any hallucinated citation,

Why does GitHub have more security layers than the Pentagon leave me alone please I don't want tokens,

Everyone: AI will soon evolve and kill us all Me: yeah in 10 years tops ppl will rediscover the finite-element method as an alternative to neural fields that works and will rebrand it as "linear splatting",

Long time no see

En route to @eurographicsc.bsky.social 2026

The neighborhood I'm living in is called "The Grange" which is literally Lagrange so I'm a Lagrangian,

We will give with Cyprien Plateau-Holleville a course on Fluid Simulation with Optimal Transport: New and Old at: - Eurographics 2026 (May 4-8, Aachen) - SIGGRAPH 2026 (July 19-23, LA) Featuring: - Gentle introduction to OT - Create your own OT at home - Fluid sim with OT See you there !

Pythagorean tiling in V&A east: www.theartnewspaper.com/2025/11/05/o...

ppl who talk politics on LinkedIn "delighted to share my recent LLM opinion! Looking for my next curated rage challenge",

Is it Chametz if Gaussian curvature >0,

A "position paper" is very Lagrangist sorry I only do (position,velocity) papers,

This is an unusual form of seamless tiling where all edges are continuous but using mirror reflection jumps, (Saughton rose garden, Edinburgh)

Happy to share a major milestone: after years of development, we are officially launching Version 1.0 of the GeometricKernels library! To top it off, our accompanying paper has just been published in JMLR (MLOSS)! 🎉 github.com/geometric-ke...

SIGGRAPH submission guidelines

Trivial (?) thought of the day: the multiplication table demonstrates that multiplication is basically integrated addition, since you can get to the same number by following any path. Or showing that (y,x) is the gradient of xy (and differential in the discrete integers),

It's funny how we give importance to putting stuff in the rebuttal as authors that we don't care about at all as reviewers. The best example is setting a positive mood in the sense of "reviewers agreed we are actually quite amazing despite criticism" lol no don't tell me what I agreed to,

Teaching curvature today. Both the tube and the single Pringle are important,

Sorry I'm Curling-free talk to me when they do diverging,

Maybe I'm the worst boomer but I don't understand the point of virtually exploring 3D environments that don't reproduce as you come back to the same point,

My Teddy's thrawn as thrawn can be an winna pey nae heed tae me,

The topology lecture of the year.

CVPR daily early morning email title: YOUR SUBMISSION IS GETTING DESK REJECTED AND DOXXED AS WE SPEAK URGENT ACTION NEEDED Content: please don't forget the review deadline in a few days,

After years (!) of work, Directional 3.0 (avaxman.github.io/Directional/) is finally released!

The response "oops haha we fixed the reviewer leak right away but listen here god help me kids if you abused it we will reject all your papers from here to eternity" seemed totally inappropriate and tone-deaf. What about the "our ultra-expensive conferences built on your unpaid work betrayed you"?

If I had papers accepted but didn't write about them here did I truly have papers accepted,

I mean it's fun to show cases where LLMs hallucinated stuff as a sort of auto da fe to the human rationale but honestly we can do a bigger demonstration of places where ppl of power and authority (at all levels) have been tear-inducingly stupid whereas LLMs gave reasonably balanced answers,

Do you have a research question or a challenge that could benefit from expert analysis, access to advanced resources, or additional funding? Consider partnering with an MSc project at The University of Edinburgh.

Interesting visual artifact I just (think I) created: Which do you think is the wider image? *Both exactly 533 pixels wide

No.... just... no papers.cumincad.org

Instead of preparing our IPAB workshop(=seminar) about any of my recent research topics, I decided to give a lecture about Linear Least Squares and all its subtleties (the four subspaces, weak solutions, pseudoinverse etc.). I feel this is something we never quite learn upfront,

Fight the power <<< embrace the weakness,

In Cambridge for the week let's avoid each other up,

Grant rejections

Is OpenReview down? I'm getting "upstream request timeout". This is AAAI Review deadline day...

"Want to access advanced AI features"? ok there's a "no" option but where's the "NO" option,

Is there any (preferably UK-based) scientist (University or company, doesn't matter) who researches AI methods to mediate between high-skill and steep-learning-curve technology and users who might need it, but do not possess the skill, the understanding, or even the idea of what to look for?

I often find chatGPT to be more reliable than wikipedia. Especially for non-mathematical topics. History and politics Wikipedia should be avoided as they are 100% hallucinations and forced agenda,

I'll be at SIGGRAPH 2025! Please avoid me,

Urban overfitters,

Two dimensional trivial solutions can be parameterized by platitudes and plongitudes,

ppl are concerned that ChatGPT uses too many em-dashes, but what about it using too many "++x"? Who even uses this? It's an "intro to C" trope,

The really bothersome thing about the promptgate is that LLMs don't automatically know to ignore this simple hack even in the context of "review this paper" command,

I think one of the things that more "tradionalists" (like me?) struggle with is that ML ppl may not consider the training of the model to be the "methodology", but rather the inference. Like they train a model and then the trained model *is the algorithm*,

If AI produces slop hallucinations concealing entire wrong statements with proper phrasing what's wikipedia's excuse,

Degenerative AI,

Who the hell called it AI Animation instead of Slop Motion