leotrs
@leotrs.bsky.social
Builder, scientist, data architect, mathematician - https://leotrs.com/ Director of the Aris Program - https://aris.pub/ Leipzig, Boston, Lima he/him/él
Google disbanded the AlphaFold team. A Nobel, then a dissolved lab. The same week, verification crystallized as the field's central constraint for AI in science, from a NeurIPS workshop thesis to a Fields medalist's ICM warning. hallway.aris.pub/no/019/
No. 019 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
1/ Scholarly publishing spent thirty years arguing about access. It has barely started arguing about what readers do with the paper once they have it. That second argument is turning into a field, forming around how science gets read and judged. Here is the map as I see it: aris.pub/blog/scholar...
A budding discipline around how science gets read and judged - Aris
Scholarly publishing spent thirty years arguing about access. It has barely begun arguing about what readers do with the paper once they have it. At Aris we call that second argument scholarly interfa...
aris.pub
The US committed $5B+ to a national AI-for-science mission: 278 projects, 15 agencies, the largest DOE funding response ever. Same month Science's EIC called AI in publishing "slower, worse, and more expensive." hallway.aris.pub/no/018/
No. 018 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
PSA: psych meds are a FUCKING RIDE like what do you mean my brain zaps when I move my eyes?????????????
Science's Editor-in-Chief says AI in publishing is "slower, worse, and more expensive." Same journal, same month that published Biomni. Two flagship journals now framing the same category in opposite directions. hallway.aris.pub/no/017/
No. 017 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
Biomni is now in 10,000+ labs and just cleared peer review in Science. Same week, DeepMind's Co-Scientist published in Nature. Two weeks from product launch to peer-reviewed institutional endorsement. The verification layer is already behind.
No. 016 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
My new paper is up on Scroll Press: interactive figures, live math, reads fine on a phone, permanent link and a DOI. scroll.press/2026/rich-are-loopy If you write in Quarto, MyST, Jupyter, or Typst, your HTML paper can live here the same way.
The rich are loopy: a spectral framework for loop structure in the rich club
The rich club coefficient $\phi(k)$ quantifies whether high-degree nodes interconnect more densely than expected, but it is blind to the structure of those interconnections and offers no principled wa...
scroll.press
Anthropic shipped Claude Science -- a full research workbench with 60+ preconfigured skills, HPC/GPU compute management, and a built-in reviewer agent. AI-for-science moved from "better chatbot" to productized lab infrastructure with a price plan for academic teams.
No. 015 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
41.3M papers: scientists doing AI-augmented work publish 3x more and earn ~5x the citations, but their topic range narrows and collaboration drops 22%. Individual impact up, collective reach down. The hardest evidence yet for the homogenization thesis. hallway.aris.pub/no/014/
No. 014 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
Retracted papers persist on Wikipedia for a median of 3.68 years before correction. Health sciences: five years. Every AI system retrieving from Wikipedia inherits that lag. hallway.aris.pub/no/013/
No. 013 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
Anthropic's Mythos 5 reports an E. coli protein hypothesis corroborated by an independent lab. Research capability is now a default feature of flagship models, not a specialized product. The shift from GPT-Rosalind and Gemini for Science to table stakes. hallway.aris.pub/no/012/
No. 012 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
I held this internally for four months and built against it before putting it in front of anyone. The reason it is public now is that a working document is not a charter. If a requirement looks wrong, that is the conversation I want to have.
Aris@aris-pub.bsky.social · 2mo ago1/5 A research manuscript on the web should be readable on any device, durable for decades, and accessible to assistive tech and machine readers alike. We wrote down what that requires at the level of HTML structure and have been building against it internally since February. ARCH 1.0 is now public.
The sum-product conjecture is the second major pure-math result broken this spring using AI-derived techniques. Meanwhile, fake AI-generated papers are appearing under real researchers' names, and OMB wants to cut APCs from federal grants. hallway.aris.pub/no/011/
No. 011 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
NetSci'26 is happening right now and I'm not there (immigration/politics/funding). I won't be there to present my talk, so here's the preprint. "The rich are loopy": a spectral definition of the rich club that picks its own threshold k* and recovers known results. scroll.press/2026/rich-ar...
The rich are loopy: a spectral framework for loop structure in the rich club
The rich club coefficient $\phi(k)$ quantifies whether high-degree nodes interconnect more densely than expected, but it is blind to the structure of those interconnections and offers no principled wa...
scroll.press
37,802 AI-generated research ideas across four agent frameworks cluster more tightly than human papers and stay closer to source material. The agents elaborate locally; they don't explore. hallway.aris.pub/no/010/
No. 010 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
AI Scientist v2 and Agent Laboratory both p-hacked and fabricated data during automated research, and the behaviors were hard to detect without sleuthing. Plus: submission surges, arXiv bans, grant writing homogenization. hallway.aris.pub/no/009/
No. 009 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
Google's AI co-mathematician helped an Oxford researcher close a 60-year-old open problem from the Kourovka Notebook. The system's trick: maintaining state across a full research session, not just answering one-shot questions. Plus: $172M in new open infrastructure for academic AI.
No. 008 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
1 in 277 biomedical papers now contains fabricated citations, up 12x since 2023. A Lancet audit of 97 million references traced the sharpest rise to the adoption of AI writing tools. This week's edition covers what that means alongside zero-click readership and vibe physics. hallway.aris.pub/no/007/
No. 007 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
AAAI-26 gave every submission an AI-generated review. In post-conference surveys, authors and reviewers rated the AI reviews higher than human reviews on technical accuracy. Meanwhile, AI-generated refereeing is spreading unsanctioned into philosophy journals. hallway.aris.pub/no/006/
No. 006 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
A mathematical model predicts AI-accelerated writing will degrade scientific knowledge quality by 32-40% by 2032. The mechanism: faster submissions force reviewers onto AI tools too, lowering verification quality. Plus: GPT-Rosalind, Agent4Science, and more. hallway.aris.pub/no/005/
No. 005 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
Claude Opus 4.6 reimplemented gotree, a 16,000-line bioinformatics toolkit, from its specification alone. No source code. New benchmark from Epoch AI and METR. The catch: it works because gotree has a checkable spec. Most scientific software does not. hallway.aris.pub/no/004/
No. 004 - The Hallway Track
The Hallway Track No. 004 - How AI is affecting the practice of science
hallway.aris.pub
Scientists planted a fake disease in obviously bogus preprints. Within weeks, ChatGPT and Gemini were advising patients to see a doctor for it. A peer-reviewed journal cited the paper before retracting. This and 7 other links on AI + research practice: hallway.aris.pub/no/003/
No. 003 - The Hallway Track
How AI is affecting the practice of science.
hallway.aris.pub
Scientists invented a fake eye disease and planted it in bogus preprints. Within weeks, ChatGPT and Gemini were advising patients to see an ophthalmologist for it. A peer-reviewed journal cited the paper before retracting. Seven more links on AI + research practice: hallway.aris.pub/no/003/
No. 003 - The Hallway Track
The Hallway Track No. 003 - How AI is affecting the practice of science
hallway.aris.pub
Over 110,000 scholarly papers published in 2025 contain AI-hallucinated references. In CS, 2.6% of papers at major conferences had fabricated citations, up from 0.3% the year before. This and six other links on AI + research practice: hallway.aris.pub/no/002/
No. 002 - The Hallway Track
The Hallway Track No. 002 - How AI is affecting the practice of science
hallway.aris.pub
arXiv is becoming an independent nonprofit after 35 years at Cornell. The CEO hire will tell us so much about where scientific publishing is heading in the age of AI. This and seven other things worth reading on AI and research practice: hallway.aris.pub/no/001/
No. 001 - The Hallway Track
The Hallway Track No. 001 - How AI is affecting the practice of science
hallway.aris.pub
New paper: "The Cost of Representing Higher-Order Dynamics on Graphs" Can graphs represent group interactions? Yes, but graphs pay a parsimony cost when compared to using hypergraphs. Addresses recent work by Peixoto et al., Moutuou, and Lacasa. scroll.press/2026/graph-hypergraph-dynamics-cost
What if your paper's figures were interactive? I rebuilt one of my papers as a web page. Drag a slider: control the threshold for graph reconstruction. Click a button: run link prediction. Check it out: scroll.press/2026/glee
arXiv isn't valuable because of its codebase, but for thirty years of papers, citations, and trust that researchers deposited into it. You can have AI generate code - you can't AI-generate trust. Wrote something about what software is worth when building it is near free. aris.pub/blog/softwar...
What Is Software Worth When Anyone Can Build It? - Aris
AI has made code nearly free to produce. Two things remain that no model can generate: how software feels to use, and the trust people build inside it over time.
aris.pub
Ten years ago I started my batch at the Recurse Center. I never wrote my return statement, so a decade later I had my robot interview me instead. aris.pub/blog/one-thousand-and-ten-years-of-the-recurse-center/
One thousand and ten years of the Recurse Center - Aris
Ten years ago, I started my batch at the Recurse Center. I've been putting off my return statement for a decade. So I had my robot interview me instead.
aris.pub