A lot of people making public judgements about math, llms, etc would benefit today from remembering that it is very hard to predict. Especially the future.
Sariel Har-Peled
@sarielhp.bsky.social
A geometer in the corn fields. https://sarielhp.org
The failure of reality to follow the analysis of smoothed analysis suggest we need a new reality, not a new type of anslysis. As no new analysis would change the way lp solvers are used in practice.
Sophie Huiberts@sophie.huiberts.me · 2w agoI have been radicalized against my own work
The trump administration to disallow Chinese companies to export advanced llm models to China...
Grunbaum's inequality states that for a convex, bounded body K in R^n, let c be its center of mass. Any halfspace containing c contains at least a 1/e fraction of the volume of K. Mind blown...
Sariel Har-Peled@sarielhp.bsky.social · 3w agoUpdated chapter on the Brunn-Minkowski inequality and several interesting applications, including the isoparemeteric inequality and Grunbaum's inequality. sarielhp.org/book/chapter...
Updated chapter on the Brunn-Minkowski inequality and several interesting applications, including the isoparemeteric inequality and Grunbaum's inequality. sarielhp.org/book/chapter...
sarielhp.org
📚 New in arXiv cs.CG: How to Catch $k$ Grid Points
How to Catch $k$ Grid Points
New paper in cs.CG (Computational Geometry)
arxiv.org
📚 New in arXiv cs.CG: The Quick Dog Jumps the Log
The Quick Dog Jumps the Log
New paper in cs.CG (Computational Geometry)
arxiv.org
Your car is spying on you, and sends the data [location, acceleration, etc] (potentially every few seconds) using its built-in SIMs. True of all new cars starting in 2018. www.bbc.com/future/artic...
Trillions of miles of data: Your car is spying on you, and it's only just the beginning
From your weight and facial expressions to your destination, cars collect a startling amount of data about you. Some of it may even raise your insurance costs.
bbc.com
SODA 2027 CFP (www.siam.org/conferences-...) absurdly encourages using a citation style (to avoid using etal) that is in contradiction to *all* writing manuals: sarielhp.org/misc/blog/26... .
SODA27 Submissions | SIAM
Submissions for ACM-SIAM Symposium on Discrete Algorithms (SODA27)
siam.org
OpenAI is doing an IPO and their technology is impressive. But is it useful? I still have serious doubts about their long term viability compared to say google. I am also wondering if small models might be able to do most targeted tasks equally well, undercutting the whole craze for data centers.
Sariel Har-Peled@sarielhp.bsky.social · 3mo agoMathematicians rank the hardness of a problem, and thus its importance, by how long it was open and how many people tried to solve it. Its a very strange norm if you think about it....
Mathematicians rank the hardness of a problem, and thus its importance, by how long it was open and how many people tried to solve it. Its a very strange norm if you think about it....
Clément Canonne@ccanonne.github.io · 3mo agoThe OpenAI breakthrough on the unit distance problem seems genuinely impressive to my semi-layman eye. (It is!) One possible takeaway, though, is that if you throw an incredible amount of money and resources at focused research, there WILL be impressive progress. I wish we tried that, too.
Wow. That is big.
Timothy Gowers@wtgowers.bsky.social · 3mo agoOpenAI's claim that this is a central conjecture in discrete geometry is not an exaggeration. This will I think be looked back on as the first time that AI solved a major mathematics problem (defined as a problem that all experts in some subfield had thought about). openai.com/index/model-...
Assume you have a discrete distribution, with i having probability p_i, for i=1,..., n, and you have access to a uniform distribution (say on [0,1]). How do you sample from this distribution in constant time? A: The alias method... sarielhp.org/misc/blog/26...
sarielhp.org
Wordle like game, but for arbitrary long words: sarielhp.org/misc/blog/26...
Pantaludounboundiglossia
sarielhp.org
There are two computers connected by two wires to each other (via two ports). For example, A1<->B1 and A2<->B2 means that port 1 of A and B are connected by a wire, as well as port 2 of both computers. Alternatively, A1<->B2 and A2<->B1. The two computers are perfectly synchronized.
Gemini suggested "Declarative Programmer" as a title for somebody who just programs using Gemini (it is really possible using gemini-cli nowadays). Other suggestions are "Declarative Programmer" and "Gemineer".
📚 New in arXiv cs.CG: The Prophet and the Voronoi Diagram
The Prophet and the Voronoi Diagram
New paper in cs.CG (Computational Geometry)
arxiv.org
www.theverge.com/science/9187... What a miserable time we've created for ourselves.
Trump fires the entire National Science Board
Federal funding for scientific research was already in turmoil.
theverge.com
Former NSF theory program director Peter Brass guest posts on the state of the foundation.
Guest Post from Peter Brass, Former NSF Theory Director, on the NSF budget.
Guest post from Peter Brass, Former NSF Theory director (though not affiliated with the NSF now) on the White House NSF budget for FY 2027....
blog.computationalcomplexity.org
It is amazing that you can generate code that generates the following image, by just chatting with an AI agent (gemini-cli in this case)... sarielhp.org/misc/blog/26...
sarielhp.org
The Dunning-Kruger-LLM effect is a cognitive bias where people with limited competence in a particular domain overestimate their own abilities since they asked an LLM to explain the domain to them.
They are eating the ducks, they are eating the goldfish... www.oxfordstudent.com/2026/02/20/p...
Péter Magyar: they are eating the ducks! – The Oxford Student
OxStu's Ivett Berényi comments on Péter Magyar's 2026 election campaign and the political situation in Hungary.
oxfordstudent.com
