I'm hiring Apache #Kafka and Apache #Flink experts (esp. on the ops side) and engineers experienced with Python for the Kafka & Flink team in @aiven_io. Helsinki 🇫🇮, Berlin 🇩🇪 and also remote in some European 🇪🇺 countries. RT please 😉

Meet async-profiler 4.2: Method tracing, Process sampling, better stacks. Now included in Amazon Corretto. github.com/async-profiler/as…

John Carmack explains how he applies Nassim Taleb's "anti-fragile" concept to his work, enjoying the thrill of new ideas while accepting that many won't succeed. Source: Deep Thoughts Engineering Speaker Series: John Carmack

A friendly reminder that incandescents are the healthiest lights you can buy. Rich in warm and biologically important near infrared light. No pulsed flicker. Purely analog. With the long night approaching, it is officially incandescent season.

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI. It weighs ~8,000 lines of imo quite clean code to: - Train the tokenizer using a new Rust implementation - Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics - Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use. - SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval) - RL the model optionally on GSM8K with "GRPO" - Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI. - Write a single markdown report card, summarizing and gamifying the whole thing. Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc. My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved. Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

🤝JetBrains and Zed are joining forces to advance the Agent Client Protocol (ACP) – an open standard that lets any compatible AI coding agent work inside any editor. 🔗 Read more: jb.gg/c9mdxr

We are open-sourcing OpenZL, a new data compression library and training tools to generator specialized compressors for structured data, achieving performance levels inaccessible to classic generic algorithms: github.com/facebook/openzl

A behind the scenes look at the making of the 1989 video game "Prince of Persia". Designer Jordan Mechner used rotoscoping to animate the movements of the game's characters, tracing video footage of his younger brother running and jumping (as well as video from old Errol Flynn films).

Blog: Method Tracing feature in async-profiler and in JFR github.com/async-profiler/as…

More broadly, I don't think a single definition of 'durable' (as in ACID D) for transactions is particularly useful. Much more useful is to ask "what kinds of failures could cause committed transactions to be lost?"

we are so back

Facebook once bought a VPN app for $120M and turned it into a surveillance tool that spied on 33M+ users' entire phones for years. This app helped Zuck buy WhatsApp for a whopping $19B and break Snapchat's encryption. Thread

See @mitchellh's AI coding workflow in action! Tomorrow at 3pm EST, we're exploring his recent PRs and discussing where agentic engineering is headed—the wins, the gaps, and the messy middle. Live Q&A included! Sign up or add the event directly to your calendar: zed.dev/agentic-engineering

Windows 98 Plus: Mystery

A big problem with debugging bugs is that the more interesting ones are hard to reproduce. That's why people build deterministic simulators to test their software or use tools like Antithesis. But why are interesting bugs so hard to trigger? 1/

The gym bros were right

A fascinating article on mitochondria. "That night, a graduate student alone in a dark laboratory in Newcastle upon Tyne in England, I became a mitochondriac: hooked on mitochondria." scientificamerican.com/artic…

I wrote about how spaced repetition systems have gotten way better over the last couple of years, thanks to the magic of ✨ machine learning ✨.domenic.me/fsrs/

I spoke on another podcast about KIP-1150 (diskless topics in Kafka). Thanks @BdKozlovski and @gharris1727 !

I designed Dropbox's storage system and modeled its durability. Durability numbers (11 9's etc) are meaningless because competent providers don't lose data because of disk failures, they lose data because of bugs and operator error. Yes S3 has lost data. No it wasn't because some disks failed. If you're building your own infrastructure you should heavily invest in release process and validation testing (link in reply). You're not going to do a better job than a major cloud provider though. The best thing you can do for your own durability is to choose a competent provider and then ensure you don't accidentally delete or corrupt own data on it: 1. Ideally never mutate an object in S3, add a new version instead. 2. Never live-delete any data. Mark it for deletion and then use a lifecycle policy to clean it up after a week. This way you have time to react to a bug in your own stack.

Today, we’re announcing the preview release of ty, an extremely fast type checker and language server for Python, written in Rust. In early testing, it's 10x, 50x, even 100x faster than existing type checkers. (We've seen >600x speed-ups over Mypy in some real-world projects.)