Louis (@LCB0B) header
Louis (@LCB0B) profile picture
Joined July 2015
Photo by Louis (@LCB0B)Photo by Louis (@LCB0B)Photo by Louis (@LCB0B)Photo by Louis (@LCB0B)Photo by Louis (@LCB0B)Photo by Louis (@LCB0B)Photo by Louis (@LCB0B)Photo by Louis (@LCB0B)Photo by Louis (@LCB0B)
Centuries before social media thirst traps, if a woman wanted to catch a suitor’s attention, she’d commission an artist to paint a portrait. Now, a team of scientists is analyzing these historic glamour shots to track shifts in female beauty standards over 300 years. The researchers’ findings reveal a kind of evolution in aesthetic preferences: Between 1650 and 1950, women portrayed in Western European portraits gradually became less curvy and youthful. Learn more: scim.ag/4tacdEH
1
9
2
16
4,864
Louis retweeted
Exquisitely detailed, endlessly ripe for experimentation, and uncompromisingly engineered, Grand Theft Auto 6 is the single most impressive video game we've have ever played. Our review: bit.ly/4ABG7ns
566
954
250
14,707
771,760
Petition style changes with AI writing support but outcomes do not improve dlvr.​it/TVvnFg
2
6
1,515
Reading all those recent beautiful essays by Terry Tao, Kevin Buzzard and @wtgowers, I could not resist, but write up my deeply personal perspective on how I perceive and see things: docs.google.com/document/d/e…
73
287
91
1,732
522,161
RIP em dash
We analyzed how @claudeai’s Opus 5.5 writes compared with Opus 5 across high-reasoning Text Arena outputs. 10 of 12 writing measures moved in a better direction. Opus 5.5 should be easier to read: - Long content words fall from 41.7% to 38.6%, the lowest share of any Claude model we analyzed. - Sentences are 17% shorter on average, dropping from 12.14 to 10.03 words. The tradeoff is length. Answers get 6% wordier, rising from 453 to 481 words on average, making Opus 5.5 give the longest answers across the Opus family. It also sounds less recognizably AI on two familiar tells: - 95% fewer em dashes - 73% fewer semicolons But a new giveaway may be emerging. Hedges and caveats such as “perhaps” and “arguably” rise 97%, from 0.39 to 0.77 per 1,000 words, the highest rate of any Claude model we analyzed. What do you think: does Opus 5.5 read more naturally?
170
410
118
13,657
1,397,860
Louis retweeted
One of the most-discussed PNAS articles on Reddit last month was “Cultural evolution of beauty standards.” Explore the article: ow.ly/EOgH50ZfmoO
1
1
20
8,177
ファッション業界は、多様性が進んだと言われる。25年分・約79万件を分析では、人種や出身国の多様性は確かに増加。 しかし、理想とされる体型はほぼ変わらず、今も痩せ型が中心。多様に見えることと、美の基準が変わることは違う。 美の基準が変わっていないことが、今の痩せ願望にもつながっているのだろう
1
10
41
2,712
New ICML 2026 paper challenges the Platonic Representation Hypothesis: model width and depth mechanically inflate similarity scores, creating a misleading global convergence trend which disappears after that bias correction. What survives is local neighborhood alignment across image, text, and video models (similar things stay near each other even across very different architectures). They call it the Aristotelian Representation Hypothesis :) So while two models may not share the same representation space in any strong global sense, they can still agree on local neighborhoods (what is similar to what). That is probably the part we actually use in retrieval, transfer, and multimodal systems, and which could be transferred between learned approximations of different models. arxiv.org/pdf/2602.14486
20
126
16
818
72,068
The speedup isn’t just in volume. On open-ended coding problems where answers are unclear, Claude’s success rate is now 76%—a 50 point jump in just 6 months. Many engineers also say Claude’s code quality is now on par with human code; we expect it to be better within the year.
44
111
62
2,201
512,008
25年分のファッションショーなどのデータを分析すると、モデルの体型の多様性は広がっているがそれは非白人のモデルに偏っており、モデルの体型の中央値は変化していないらしい。
1
3
633
Louis retweeted
Closed labs hide model sizes. They can't hide what their models know, and what a model knows is an indicator on how big it is. Reasoning compresses. Factual knowledge doesn't. So you can size a frontier model from black-box API calls alone, and across releases you can literally watch a single fact arrive in the parameters over time. For three years, my friends Jiyan He and Zihan Zheng have been asking frontier LLMs the same question: "what do you know about USTC Hackergame?", a CTF contest. May 2024: GPT-4o invented fake titles. Feb 2025: Claude 3.7 Sonnet listed 19 verified 2023 challenges. By April 2026, frontier models recall specific challenges across consecutive years. After DeepSeek-V4 dropped, I instructed my agent to spend four days autonomously turning that habit into Incompressible Knowledge Probes (IKP) — 1,400 questions, 7 tiers of obscurity, 188 models, 27 vendors. Three findings: 1/ You can approximately size any black-box LLM from factual accuracy alone. Penalized accuracy is log-linear in log(params), R² = 0.917 on 89 open-weight models from 135M to 1.6T params. Project closed APIs onto the curve → GPT-5.5 ~9T, Claude Opus 4.7 ~4T, GPT-5.4 ~2.2T, Claude Sonnet 4.6 ~1.7T, Gemini 2.5 Pro ~1.2T (90% CI: 0.3-3x size). 2/ Citation count and h-index don't predict whether a frontier model recognizes a researcher. Two researchers with similar citation profiles get very different responses. Models memorize impact — work that shaped a field, not many incremental papers. 3/ Factual capacity doesn't compress over time. Across 96 open-weight models across 3 years, the IKP time coefficient is statistically zero, rejecting the Densing-Law prediction of +0.0117/month at p<10⁻¹⁵. Reasoning benchmarks saturate; factual capacity keeps scaling with parameters. Website: 01.me/research/ikp/ Paper: arxiv.org/pdf/2604.24827
70
227
55
2,158
394,754
🚨Very excited to see our work on warmth & sycophancy in LLMs out in @Nature today!🚨 We study what happens when LLMs are fine-tuned to be warmer, and find that warmth and sycophancy can be linked, with warm models showing higher errors on a range of benchmarks (🔗s below)
15
62
11
268
38,230
With max thinking Opus 4.7 is quite impressive, with a real sense of style In two prompts: "implement the Tower of Babel, in 3D, in as sophisticated and visually interesting a way as possible. It should be interactive" and then "make it better." Play: tower-of-babel-1776392618.ne…
31
50
9
770
93,775
If you're wondering whether saturating ARC-AGI-1 or 2 means we have AGI now... I refer you to what I said when we launched ARC-AGI-2 last year (which is also the same thing I said when we announced ARC-AGI-2 was coming, in Spring 2022, before the rise of LLM chatbots)... The ARC-AGI series is not an AGI threshold, it's a compass that points the research community toward the right questions. ARC-AGI-1 is a minimal test of fluid intelligence -- to pass it, you needed to show nonzero fluid intelligence. This required AI to move past the classic deep learning / LLM paradigm of pretraining scaling + static models at inference, toward test-time adaptation. ARC-AGI-2 is the same, but with tasks that probe deeper levels of reasoning complexity (particularly with regard to concept composition). Still, these are tasks that are solvable in minutes by regular people with no external tool use (we hired our test takers off the street), so it does not represent the upper bound of what human fluid intelligence can achieve (say, solving a Millennium problem). ARC-AGI-3 (launching March 2026) probes interactive reasoning: we evaluate how systems explore unknown environments, model them, set their own goals, and plan/execute towards these goals, autonomously, without instructions. We have also started work on ARC-AGI-4 and ARC-AGI-5, which I am pretty excited about!
90
88
22
1,067
212,520
The tiny Caribbean island of Anguilla now derives almost half of its state budget from the sale of .ai domain names.
94
551
90
10,061
479,794
An implementation of Contrastive Poincaré Maps with sample notebooks for the biomedical case studies is now publicly available: bit.ly/3I0sM2J
Single-cell data can reveal hierarchical patterns in organismic development but popular embedding approaches often distort them. We introduce Contrastive Poincaré Maps, a self-supervised hyperbolic encoder that preserves hierarchies, scales efficiently, and uncovers developmental lineages across diverse datasets. Led by @nithyabhasker. Preprint here: bit.ly/4211hMY
4
15
1,331
Single-cell data can reveal hierarchical patterns in organismic development but popular embedding approaches often distort them. We introduce Contrastive Poincaré Maps, a self-supervised hyperbolic encoder that preserves hierarchies, scales efficiently, and uncovers developmental lineages across diverse datasets. Led by @nithyabhasker. Preprint here: bit.ly/4211hMY
18
3
69
16,237
In this article, Boucherie et al. apply physics-based models to the arrangement of locations to study how geography shapes human movement. They find an underlying pattern in how people choose to move, independent of geographical layout. nature.com/articles/s41562-0…
4
19
2,224