A few weeks ago, I tried to explain to someone what Microsoft Copilot is. I couldn’t… because the name ‘Copilot’ now refers to at least 75 different things.

Can a large language model (LLM) improve at code generation using only its own raw outputs, without a verifier, a teacher model, or reinforcement learning? We answer in the affirmative with simple self-distillation (SSD): sample solutions from the model with certain temperature and truncation configurations, then fine-tune on those samples with standard supervised fine-tuning. SSD improves Qwen3-30B-Instruct from 42.4% to 55.3% pass@1 on LiveCodeBench v6, with gains concentrating on harder problems, and it generalizes across Qwen and Llama models at 4B, 8B, and 30B scale, including both instruct and thinking variants. To understand why such a simple method can work, we trace these gains to a precision-exploration conflict in LLM decoding and show that SSD reshapes token distributions in a context-dependent way, suppressing distractor tails where precision matters while preserving useful diversity where exploration matters. Taken together, SSD offers a complementary post-training direction for improving LLM code generation.

Legal intern Raj Gambhir was the principal author of this post.The Trump administration has restricted the First Amendment right to record law enforcement by issuing an unprecedented nationwide flight restriction preventing private drone operators, including professional and citizen journalists,...

Did hell freeze over? Not quite: the driver belongs to Tiny Corp, not Nvidia, you’ll have to compile it with Docker rather than plug-and-play, and it’s designed for LLMs. But you no longer need to disable Apple’s System Integrity Protection (SIP), because Apple is letting that driver get signed, Tiny says.
[Image: Apple finally approved our driver for both AMD and NVIDIA,” Tiny writes. https://platform.theverge.com/wp-content/uploads/sites/2/2026/04/HEy9-blbAAABy-J.jpg?quality=90&strip=all]

Claude Code has gotten extremely good at finding security vulnerabilities, and this is only the beginning.

llm-wiki. GitHub Gist: instantly share code, notes, and snippets.

As online betting has grown in popularity, a new report from the New York Federal Reserve builds on the troubling link between legal sports wagering and financial health.

Rootless Podman for Android — run Linux containers on your phone - ExTV/Podroid

Interpretability research from Anthropic on emotion concepts

Iranian strikes are taking out AWS servers in the region.

Shared LLM access via cohort subscriptions

submit your favorite sites!

Welcome to the era of sprawling, idiosyncratic tooling.

We run a production Rails store on SQLite — not Postgres, not MySQL. A single file on a Docker volume. It works surprisingly well until two containers try to...

TurboQuant WASM SIMD vector compression — 3 bits/dim with fast dot product. Requires relaxed SIMD (Chrome 114+, Firefox 128+, Safari 18+, Node 20+) - teamchong/turboquant-wasm

NFC hackathon badges for the Overglade game jam in Singapore! - KaiPereira/Overglade-Badges

I saw a post on LinkedIn the other day from a self-proclaimed 20 year agency veteran of WordPress saying that was it, they’re moving the entire agency off of WordPress and onto AI. Now, because I, too, am a 20 year veteran of WordPress, this kind of story catches my attention. He posted that they […]

A Racket programming environment for your iPhone and iPad. Write, run, and explore Racket code on the go.