Sign in

Dimitri Lozeve

@dimitri.lozeve.com
39 followers 108 following 23 posts

Research engineer working on #NLP models for law. Here for discussions about #mathematics, #stats, #MachineLearning, #nlp, #optimization... lozeve.com

PostsRepliesMedia
Reposted by Dimitri Lozeve
Nobel Prize @nobelprize.org · 13h
BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2026 #NobelPrize in Physics to Francis Halzen “for decisive contributions to the IceCube Neutrino Observatory and the discovery of high-energy neutrinos of astrophysical origin.”
9463242
Reposted by Dimitri Lozeve
Nutan Limaye @nutanlimaye.bsky.social · 17h
Unbelievable stuff! arxiv.org/abs/2610.067... APSP and 3SUM refuted!! This is big.
arxiv.org
Truly Subquadratic 3SUM and Truly Subcubic APSP via Triangles in Sparse Lopsided Graphs
We give the first polynomial improvements over the textbook algorithms for $3$SUM and All-Pairs Shortest Paths (APSP): we show how to deterministically solve $3$SUM on $n$ integers of polynomial size ...
23511
Dimitri Lozeve @dimitri.lozeve.com · 05/10/2026
The official "popular information" article from the Nobel Prize committee is a very good exposition for non-specialists: www.nobelprize.org/prizes/medic...
Figure 4 from the linked article
000
Dimitri Lozeve @dimitri.lozeve.com · 04/10/2026
It's also very interesting to read about the engineering work in optimizing and simplifying these engines. And the incentives behind them: wanting to reuse rocket engines make them think very differently about reliability for instance
000
Dimitri Lozeve @dimitri.lozeve.com · 04/10/2026
This article by Brian Potter on SpaceX's new version of their Raptor engine is also a great primer on rocket engine design: www.construction-physics.com/p/how-spacex...
construction-physics.com
How SpaceX Streamlined the Raptor Engine
If you’re reading this, there’s a very good chance you’ve seen this famous image of three iterations of SpaceX’s Raptor rocket engine.
100
Reposted by Dimitri Lozeve
Matt Henderson @matthen.com · 03/10/2026
Visualising the distribution of substrings in the genomes of different organisms. Going up to depth eight by recursively splitting into quadrants, we get these fractal-like pictures.
816634
Dimitri Lozeve @dimitri.lozeve.com · 02/10/2026
The importance of having a proper hypothesis, causal model, and controls...
130
Dimitri Lozeve @dimitri.lozeve.com · 02/10/2026
A study showed that nuclear power plants were associated with higher incidence of cancer. By reproducing the study, researchers showed that the statistical methods used led to higher cancer incidence around any landmark: www.breakthroughjournal.org/p/things-tha...
breakthroughjournal.org
Things that Apparently Cause Cancer
What does this methodology measure anyway?
1145
Dimitri Lozeve @dimitri.lozeve.com · 01/10/2026
Google has released WeatherNext 3, their new weather model: blog.google/innovation-a... It is interesting because the AI model does not rely solely on numerical simulations: it incorporates live data from satellites, which apparently increases the accuracy for fast-changing meteorological events
Figure from the article: The end-to-end WeatherNext 3 system architecture
000
Dimitri Lozeve @dimitri.lozeve.com · 30/09/2026
Very well-written essay: "Applied mathematics has met the machine before" proofsandprompts.com/2026/09/28/a...
proofsandprompts.com
Applied mathematics has met the machine before
Why does AI worry applied mathematicians less? We have met the machine before: human computers moved up to designing algorithms, and a computer run led to the solution. Driven by problems rather th…
001
Dimitri Lozeve @dimitri.lozeve.com · 23/09/2026
Social scientists and computer scientists need to work together to understand and manage this change. The AGI "boomer/doomer" debate is a distraction on this discussion of actual impact.
000
Dimitri Lozeve @dimitri.lozeve.com · 23/09/2026
The real question is not whether LLMs are on a path to "AGI", but what are the likely social consequences of LLM deployment: cultural homogenization vs. diversity, who wins and loses, and power concentrating in whoever owns the models.
100
Dimitri Lozeve @dimitri.lozeve.com · 23/09/2026
LLMs in particular can compress a lot of information into a single artefact, and allow to remix it at large scale and speed (similar to the price signal in a market).
100
Dimitri Lozeve @dimitri.lozeve.com · 23/09/2026
All technological innovations (like electricity or the Internet) can have large cultural impact, but cultural technologies are peculiar in that they operate on our society and culture directly.
100
Dimitri Lozeve @dimitri.lozeve.com · 23/09/2026
The authors interpret LLMs as a socio-cultural technology, like the printing press, markets, or representative democracy.
100
Dimitri Lozeve @dimitri.lozeve.com · 23/09/2026
📄 Large AI models are cultural and social technologies www.science.org/stoken/autho... I found this article very interesting, as a counter-narrative to the common description of LLMs as abstract entities with vague definitions ("agents", "AGI", etc) ⤵️
Screenshot of the first page of the paper
100
Dimitri Lozeve @dimitri.lozeve.com · 14/09/2026
Sometimes there are problems that seem so "basic" that they should be essentially solved And then someone finds a way to dramatically improve the performance In this case, conversion from Unix epoch to year-month-day: www.benjoffe.com/fast-date
benjoffe.com
The Julian Map: A Faster technique for Gregorian Date Conversion
In this article I show step-by-step how the fastest date algorithm can be sped-up by a further 2-12%.
000
Dimitri Lozeve @dimitri.lozeve.com · 14/09/2026
So AI usage has short-term benefits, but in the long term, it mainly benefits workers who are already experienced. For junior workers, it probably depends on _how_ they use AI, and whether they use it to learn and grow instead of just doing their work for them
000
Dimitri Lozeve @dimitri.lozeve.com · 14/09/2026
- After three months, lawyers who had access to AI outperformed controls, but only for people that were already experienced. Junior lawyers were split into two categories: those that greatly outperformed the controls and those that greatly underperformed them, with very few in-between.
100
Dimitri Lozeve @dimitri.lozeve.com · 14/09/2026
Their findings: - AI usage raises the quality of work output, especially for junior workers
100
Dimitri Lozeve @dimitri.lozeve.com · 14/09/2026
"Does AI Assistance Enhance or Erode Expertise? Evidence from a Three-Month Field Experiment in Patent Drafting" www.nber.org/papers/w35720 New paper by David Autor et al, that studies the impact of AI usage on professional expertise with a pre-registered statistical study, on patent lawyers ⤵️
Figure 3 from the paper: Performance Distributions of Treated and Untreated Subjects on Redlining Task: Overall and by Experience Level
100
Reposted by Dimitri Lozeve
John C. Baez @johncarlosbaez.bsky.social · 10/09/2026
The moral culpability rests on people here, even if those people want to hide behind LLMs or trillion-dollar companies. We read about the team of ten at OpenAI who prodded Astra to cough up the Navier-Stokes paper... but they are conveniently anonymous.
2191
Reposted by Dimitri Lozeve
Gro-Tsen @gro-tsen.bsky.social · 10/09/2026
So it's getting clear that some of the math abilities AI models have become ❝superhuman❞ at is “not giving credit where credit is due”, like “failing to cite previous work on the subject that inspired you” and, worse, “stealing colleagues' ideas”. 🧵⤵️
34921
Dimitri Lozeve @dimitri.lozeve.com · 09/09/2026
Several times I've seen the Langlands program mentioned in passing, like "a set of links between different areas of mathematics", but I've never read a clear description. This article in @quantamagazine.org does a good job explaining it with specific examples: www.quantamagazine.org/what-is-math...
quantamagazine.org
What Is Math’s Mysterious Langlands Program Really About? | Quanta Magazine
Hidden correspondences hint at a deeper structure to the mathematical universe. Our columnist unpacks one of those connections and asks mathematicians what they might mean.
000
Reposted by Dimitri Lozeve
lynn @chordbug.bsky.social · 24/06/2026
fixed your headline
A hit song on the Billboard 100 charts is termed *$k$-good* if it is _ever_ in a weekly top-$k$, and *$k$-immediate* if it _debuts_ in the top-$k$.


An artist is *$(n,k)$-Olivian* if their $n$ first $k$-good hits are all also $k$-immediate.

*Headline 1.* Olivia Rodrigo is the first $(10,10)$-Olivian woman.
392041558
Dimitri Lozeve @dimitri.lozeve.com · 04/06/2026
If you want some positive, hopeful news, there are some incredible recent advances in cancer research: www.abundanceandgrowth.org/p/cancer-bre...
Survival curves comparing daraxonrasib versus chemotherapy for patients with pancreatic cancer with RAS G12 mutations. Credit: NEJM. (From the linked article)
010
Reposted by Dimitri Lozeve
elisa freschi @elisafreschi.bsky.social · 28/05/2026
"there is a real, identifiable competence that the humanities, done well, build in a person […]You can only get it by spending ys reading difficult texts, being forced to defend yr interpretations of them in writing & being told repeatedly that you have not yet made yr argument as well as you could"
1104
Dimitri Lozeve @dimitri.lozeve.com · 28/05/2026
"Bureaucracy" is often used as a bad word. But this paper argues that formalization can take two very different directions: "coercive" or "enabling" (guess which one corresponds to the stereotype!) The intuition is that rules are welcome when designed to empower people! lozeve.com/posts/types-...
lozeve.com
Dimitri Lozeve - Two Types of Bureaucracy: Enabling and Coercive
Dimitri Lozeve's blog: Two Types of Bureaucracy: Enabling and Coercive
010
Reposted by Dimitri Lozeve
Gro-Tsen @gro-tsen.bsky.social · 23/05/2026
Very insightful short essay by Petra Schwer ( mathstodon.xyz/@PetraSchwer ): ‘The meaning of doing mathematics’ “(Can AI solve all math? What do we actually mean by doing mathematics? How do we communicate mathematics? What is mathematics beyond problem solving?)” arxiv.org/abs/2509.15998
arxiv.org
The meaning of doing mathematics
Can AI solve all math? What do we actually mean by doing mathematics? How do we communicate mathematics? What is mathematics beyond problem solving? This essay is my attempt to answer these question...
22712
Dimitri Lozeve @dimitri.lozeve.com · 22/05/2026
If the universe is deterministic, can we have free will? The "compatibilism" debate is important in #philosophy, but there have also been surveys of what people around the world believe! #paper: philpapers.org/rec/CHAIBI-2
philpapers.org
Hagop Sarkissian, Amita Chatterjee, Felipe de Brigard, Joshua Knobe, Shaun Nichols & Smita Sirker, Is Belief in Free Will a Cultural Universal? - PhilPapers
Recent experimental research has revealed surprising patterns in people's intuitions about free will and moral responsibility. One limitation of this research, however, is that it has been conducted e...
030
Dimitri Lozeve @dimitri.lozeve.com · 21/05/2026
New blog post! I'm currently exploring all kinds of #scheduling algorithms, and I stumbled upon "Johnson's Rule". It's a very simple way to find the optimal ordering of a series of tasks that need to go through two steps (machine A and machine B) in order: lozeve.com/posts/johnso... #optimization
The optimal schedule of a random project, computed with Johnson's Rule
010