Topic
Everything filed under Research, newest first.
RSS · JSON · All topics
Automatic Checking of Compiler Correctness with E-Graphs
MIT professor Adam Chlipala describes CSAIL research on using e-graphs to automatically check that compiler optimizations are correct, as part of his series on building correct, efficient programs from proven rewrite rules.
13 min · 2,942 words
What actually makes a dog or cat live longer? We ranked the evidence.
Fluffmobile reads the large studies behind pet longevity advice and a 2025 GRADE-based review, ranking factors by certainty and effect size, and finds the things that matter most for dog and cat lifespan are mostly boring while some popular advice has almost no evidence.
8 min · 1,726 words
A field of physics, audited by machine
Casey Handmer had AI map 158,059 theoretical astrophysics papers around Caltech's TAPIR group, screen 50,269 full texts and recompute 659 highly cited papers, verifying 872 errors in 433 papers and tracing how often later work copied rather than corrected them.
8 min · 1,894 words
SWE-CoWork: Benchmarking Coding Agents as Digital Employees in Organizations
TokenWave AI's SWE-CoWork benchmark drops coding agents into simulated engineering orgs with 26 colleagues across 101 real repositories; the two leading models pass about 63% of target tests but complete fewer than 8% of assignments end to end.
1 min · 240 words
Please Implement A Quorum Database
Transcript of an HPTS 2026 talk arguing that leaderless, quorum-based replication is practical for production databases today and deserves more use than Raft-style leader replication, walking through the design space, prior systems, and the techniques that make quorums work.
16 min · 3,776 words
Deepnote's open benchmark drops 12 frontier models into the Slack of PageMantle, a synthetic company with 11 services, across 50 workplace situations. GPT-6 Astra leads at 8.97/10; models help well but struggle to stay silent, and pricier models aren't always better.
27 min · 6,246 words
Can a formally-verified C compiler be as performant as gcc -O2?
Aananth V describes writing about 15,000 lines of Rocq, with help from coding agents, to add verified optimization passes to the CompCert C compiler in an attempt to close its 28% performance gap with gcc -O2, and what still blocks parity.
23 min · 5,342 words
Formal Verification is Not Verifiable
Alperen Keles argues that as AI pushes formal verification mainstream, the specifications and models behind 'formally verified' claims are themselves hard to verify, and that testing's verifiability makes it the tool to falsify, harden and validate those models.
7 min · 1,535 words
Exact Gradients Through a BRep Kernel
Municipal Robotics shows how to take exact derivatives of CAD parameters, including fillet radii, through its vcad BRep kernel by freezing tessellation topology and differentiating node positions, checked against finite differences.
33 min · 7,492 words
What thousands of conversations look like inside a language model
Operator_0x builds an interactive atlas of Qwen3-1.7B's residual-stream representations, mapping thousands of conversations to see whether similar topics and the same ideas across languages land in neighboring regions of the model's internal state space.
9 min · 2,112 words
Watch a Brain Learn: A visual explanation of reinforcement learning
An interactive visual explainer that builds a simple digital creature with a brain and watches it learn a new behaviour, illustrating how reinforcement learning works in brains and machines.
9 min · 2,071 words
Advancing science & engineering with Antigravity's /teamwork
Google Research shares results from Antigravity's /teamwork multi-agent workflows, used with human experts to tackle long-open problems in mathematics, chemistry, systems software and ML engineering, and says the tools are being made available externally.
13 min · 3,076 words
Beating the frontier on NetHack
Antithesis explains how its exploration fuzzer, which can now find a path to victory within five wall-clock minutes on nearly 80% of NetHack seeds, got there, and how the same search-and-optimization approach applies to testing software.
77 min · 17,750 words
A Click-Fraud Bot Showed Me 5,000 GitHub Accounts Anyone Could Try to Reset
Investigating a GitHub Actions click-fraud campaign running on dormant accounts, Shmulik Cohen found the attacker picked old accounts registered with public disposable inboxes like Mailinator, then counted 5,552 such accounts from GH Archive data and reported the risk to GitHub.
5 min · 1,164 words
Erik Hoel proposes qualiaphobic AI: neural architectures deliberately engineered so no theory of consciousness could deem them conscious, via lookup-table anchoring, fragmenting phenomenology, and adversarial design, arguing this solves AI welfare and keeps humans the only moral patients.
6 min · 1,337 words
Technology as a Search Problem
Brian Potter frames technological progress as a search through a huge space of possible designs, using simple simulated models to explore what makes technological search easy or hard and how information-gathering methods change outcomes.
22 min · 5,006 words
Laurence Tratt describes ykifying CPython: using the yk meta-tracing system to add a JIT compiler to the large CPython interpreter, walking through optimising interpreter dispatch, tracing, unrolling and indirect inlining, and showing ykcpython can be surprisingly competitive on performance.
27 min · 6,157 words
Researchers discover rare star system that emits radio signals every eight minutes
Curtin University's ICRAR node linked long-period radio pulses to GPM J071507−114153, a white dwarf spinning every eight minutes orbiting a Sun-like G-type star — the first radio pulses detected from a white dwarf with a Sun-like companion, published in Nature Astronomy.
3 min · 592 words
CF Bolz-Tereick looks back on 20 years of Pyrolog, an RPython-based Prolog implementation begun in 2006 inside PyPy, covering its role in PyPy's JIT work, continuation-based architecture, compressed terms, Unipycation, Pycket, Effekt and a 2026 git conversion.
13 min · 3,070 words
Eval Awareness vs. Eval Paranoia
Dean Valentine, who has built alignment evals full time for six months, argues frontier models are not that eval aware, gives concrete examples, and offers a tentative explanation for behaviour that people often mistake for consistent evaluation awareness.
7 min · 1,657 words