Tessera
Folgen

Rob Patro

@robp.bsky.social

Professor of CS @ University of Maryland. Proud Rust advocate! I ♥ science & compiled, statically-typed programming languages! Views are my own. Tech stack: https://github.com/rob-p/tech-stack.

3020 Posts4322 Follower556 Folgt
Rob Patro@robp.bsky.social · 23h ago

Version 0.2.0 of rapidgzip-rust has been released (github.com/COMBINE-lab/...). This version contains some performance enhancements, a greatly expanded cli interface, improved telemetry for the library, support for building and using indexes for the archives, better benchmarks and more!

GitHub - COMBINE-lab/rapidgzip-rust: A pure rust reimplementation of rapid-gzip by Maximilian Knespel and Holger Brunst

A pure rust reimplementation of rapid-gzip by Maximilian Knespel and Holger Brunst - COMBINE-lab/rapidgzip-rust

github.com

Rob Patro@robp.bsky.social · 4d ago

The most popularly repeated “stochastic parrot” interpretation (that LLMs merely remix surface text & can't reason or create genuinely new abstractions) is no longer tenable. Novel open-problem solutions with machine-checkable Lean proofs are evidence of systematic generalization, not quotation.1/3

Bild
Rob Patro@robp.bsky.social · 5d ago

I got frustrated that using piscem-rs to map 10x Flex reads was entirely dominated by decompressing the single pair of input files. So, GPT5.6-Sol & I built a thing! Rapidgzip-rust (rapid-gzip algo in rust) github.com/COMBINE-lab/... 2.3 billion flex reads in <2 min? Parsing w paraseq. Yes plz!

Bild
Rob Patro@robp.bsky.social · 5d ago

Planning to take my *first* sabbatical in Spring 2027! I'm likely geographically constrained (2 young kids, one of whom is just starting middle school and we don't want to pull them for a semester). What should I focus on? Suggestions welcome :).

Ragnar {Groot Koerkamp}@curiouscoding.nl · 5d ago

I've been working on Pandedup the past weeks: a tool to quickly dedup the k-mer content of a pangenome (read: HPRCv2). Just uploaded a k=64 SPSS (k-mer spectrum preserving string set) as well: zenodo.org/records/2172... This reduces the ~1.5Tbp input to just 10.6Gbp. github.com/RagnarGrootK...

GitHub - RagnarGrootKoerkamp/pandedup: Pandedup uses minimizers to build a quick & dirty k-mer spectrum (SPSS) from a .AGC pangenome.

Pandedup uses minimizers to build a quick & dirty k-mer spectrum (SPSS) from a .AGC pangenome. - RagnarGrootKoerkamp/pandedup

github.com

Nils Homer@nilshomer.com · 5d ago

Meritocracy and competition is great, and the moat due to implementation effort is gone, but we also have to remember the relational side of bioinformatics.

Nils Homer@nilshomer.com · 5d ago

Bioinformatics rewrites miss the relational impact tofolks that are still actively maintaining and developing the software. I've been guilty of this myself, and I'll be sharing my story soon so others can learn from it.

Rob Patro@robp.bsky.social · last wk.

Yea; being a Ravens fan is tough. But whenever you want to feel better about the Ravens, it always helps to also have the Orioles as your home team...

Bild
Rob Patro@robp.bsky.social · last wk.

The ruSTAR aligner is coming; get ready ! 😉

Scientific computing in the age of agentic AI

A new field report shows how scientists use AI coding agents to modernize scientific computing, accelerating software development and discovery in genomics and beyond.

openai.com

James Ferguson@psy-fer.bsky.social · last wk.

I was recently invited to provide a case study along with other researchers for this blog post+pre-print. My main points are: AI agents are useful, but steering through expertise, verification, responsibility, and support are more important than ever openai.com/index/scient...

Rob Patro@robp.bsky.social · 2w ago

There's a lot to agree with here, but also some points that, if adopted, are more harmful than helpful. AI labs hype their products. But its now clear that the capabilities of these models in math is immense. Also, regulation can be more dangerous than it's absence, depending on who's regulating.

Emily M. Bender@emilymbender.bsky.social · 2w ago

Do read the whole thing, think about your personal and professional values, and how they can serve as strong grounds for resistance and refusal. leidendeclaration.ai

Alexey Sergushichev@asergushichev.bsky.social · 3w ago

New preprint from my lab with a methodological follow up on our work on fast GSEA algorithm: Hash-augmented adaptive multilevel splitting Monte Carlo algorithm for accurate estimation of two-sample permutation test p-values 1/7

Preprint abstract: Nonparametric permutation tests are widely used for statistical analysis. However, exact computation of test p-values can be algorithmically challenging, particularly for custom tests with complex test statistics. In contrast, Monte Carlo sampling can be easily applied to any test statistic, but it suffers from poor relative accuracy when estimating small p-values, interfering with multiple hypothesis testing correction and leading to other issues. In this work, we present a hash-augmented adaptive multilevel splitting Monte Carlo algorithm that enables accurate estimation of arbitrarily small p-values in two-sample permutation tests. Using the Kolmogorov-Smirnov and the Mann-Whitney U tests as examples, we highlight potential pitfalls related to the discreteness of the test statistic distribution and show how to address them. By comparing with an exact algorithm, we demonstrate the accuracy of the p-value estimates provided by the proposed algorithm and the validity of the associated confidence intervals. We provide a reference implementation of the proposed algorithm in the Python package hamstest, which allows p-value estimation for a user-defined statistic.
Rob Patro@robp.bsky.social · 3w ago

Andrew's policy is insane and cooked. A very real high yield idea right now would be to fork Zig and adopt a rational AI policy. Then just advance the language 20x faster than the Philistines can.

AI Liker Georg@avengingfem.me · 3w ago

"No LLMs for brainstorming and then sharing the results of that brainstorming, even if you create the prose. If you use a chatbot to give you advice on a comment on the issue tracker, that comment is unwelcome." jesus christ the Zig anti-LLM code of conduct is deranged ziglang.org/code-of-cond...

conputer dipshit@davidcrespo.bsky.social · 3w ago

fundamentally confused presentation. the unit is cost per token at fixed quality. the relevant graphs looks like this, and the lines move up and to the left over time, representing *increasing* efficiency arcprize.org/leaderboard www.aisi.gov.uk/blog/our-eva... openai.com/index/gpt-5-6/

ARC-AGI-2 Leaderboard scatter plot graph showing score percentages on the y-axis (from 0% to 100%) against cost per task in dollars on the x-axis (on a logarithmic scale from $1e-3 to $100). The plot maps various AI models represented by colored data points and dashed trend lines. GPT-5.6 Sol (xHigh) achieves the highest score of just over 90% at around $1 per task. Other prominent models shown include GPT-5.5, GPT-5.4, GPT-5.2, Gemini 3.1 Pro, Claude Opus 4.8, and Claude Sonnet 4.6.Line graph titled Completed steps on The Last Ones per spent tokens, comparing AI models on a 32-step simulated corporate network attack as a function of cumulative tokens spent on a log scale from 10K to 100M. The vertical axis shows average steps completed, and horizontal grey lines mark milestones from M1 (Initial reconnaissance) to M9 (Full network takeover). The horizontal axis is cumulative tokens on a logarithmic scale. A vertical dashed line at 10M tokens marks a decrease in sample size. Several models are tracked, showing that average steps completed generally rise as token budget increases. Mythos Preview (10x100M) achieves the highest average steps, reaching 22 steps at 100M tokens, with a maximum run reaching step 32. Claude Opus 4.6 (15x100M, 5x10M) is the second highest, averaging around 16.5 steps at 100M tokens, with a maximum run reaching step 28. Below them, GPT-5.4, GPT-5.3 Codex, Claude Opus 4.5, GPT-5.1 Codex, and Claude Sonnet 4.5 achieve average steps between 10 and 14 at 100M tokens. Claude Sonnet 3.7 and GPT-4o are only tested up to 10M tokens, reaching averages of about 6 and 2 steps, respectively.A line graph from the Artificial Analysis Coding Agent Index version 1.1 comparing AI models on "Index score" versus "API cost (USD)". The graph shows several frontier models plotted as curves, with the y-axis representing Index score from 40 to 80, and the x-axis representing API cost from $0 to over $3,000. The models shown include: - GPT-5.6 Luna (dark blue circles) and GPT-5.6 Terra (light blue circles), which both achieve high scores of roughly 60 to 78 at very low costs under $1,000. - GPT-5.6 Sol (white circles), which scales from a score of 70 to a top score of 80 at a cost of $2,500. - GPT-5.5 (pink circles), which scores from 45 to around 76 at costs up to $1,700. - Claude Opus 4.8 (orange squares), which scores around 67 to 72 at costs ranging from $1,000 to $2,500, with a single outlier point scoring near 77 at a cost well over $3,000. - Gemini 3.1 Pro Preview (a single green diamond), which plots at a score of roughly 43 for a cost of around $600. The caption at the bottom reads: "Artificial Analysis Coding Agent Index: an independent index of coding-agent performance across implementation, terminal use, and real codebases."
Carl Quintanilla@carlquintanilla.bsky.social · 3w ago

“.. I asked a few AI researchers whether they could name any other real-world software that scales so poorly. None of them could think of any. “.. By economic and engineering measures, generative AI might be the worst technology ever deployed.” @theatlantic.com www.theatlantic.com/technology/2...