Sneak peak at a slide from my RCU talk at @KernelRecipes . This "Life cycle of a grace period" block diagram serves as a reference for me too now ;-)

Every single one of the 11 Meta superintelligence hires is an immigrant who did their undergrad abroad. 7 China, 1 India, 1 Australia, 1 UK, 1 South Africa. 8 are PhD or PhD dropouts in the US. Immigration is key to US AI innovation.

A bit late to the party, but I really think #GitLab is a very nice alternative to #GitHub. The user interface feels a lot simpler and intuitive, and the free repository limit of 10GB/repo is 5x that of GitHub. A kernel repo is >2GB with history. I’ll continue to use both though.

@OpenAI

Elon's week be like...

I am wondering why #LLM coding benchmarks like #Aider Polyglot reveal the coding problems they are testing on. Doesn’t that give the models an opportunity to be trained on the problems in advance, thus defeating the scoring? https://t.co/UBgKS9eiJo https://t.co/hDGwJsHc93

It is my honor to be designated as Godfather of the 2025 Kernel Recipes. The conference organizers asked me to help increase the speaking skills of the presenters. This blog series is a first step in that direction:

This is safety beyond the usual unsafe memory access protections and is more than welcomed and fascinating! Especially since issues are caught much earlier at compile time! 2/

I’m fascinated that rust lang couples synchronization primitive types with structures they protect thus allowing for common programming mistakes to be caught at compile time! (ex, dropping a lock too early or holding wrong lock like rwlock in read mode during mutations). 1/

Here's our statement on the results of the 2024 presidential election:

AI video gen is heating up. This is going to disrupt commercial video advertising AFAICT. Multiple ready to use tools now exist: Kling ai, Flair ai, runway ML etc

https://t.co/RCsqQXBnNE OpenSSL adds a number of new encryption algorithms, support for "“directly fetched composite signature algorithms such as RSA-SHA2-256”"

The ourouborous has begun. Wikipedia is an important source of training data for AIs. At least 5% of new Wikipedia articles in August were AI generated (To be clear, this does not mean that AI will fail as it trains on its own data, synthetic data is already a part of training)

Half the country is voting for a climate denying criminal fascist. This is absolutely insane #Harris

No one should ever blindly trust anything coming out of a large language model & IMO avoid using it for anything beyond simple / pattern-matchy tasks: https://t.co/eaYxuB3LdW AI including new OpenAI models (o1) lack surprisingly by a staggering amount in reasoning, as paper finds

If you’re going to scheme, scheme with the best people in your field!

My main motivation for attending conferences is to get fired up about stuff and meet people. This time for #LPC, I had a lot of fun digging into sched_ext internals with some talented people. As for attending presentations, those are recorded, so I'm not sweating it too much.

TIL, Austria is not a NATO member.

Our teachers, nurses, and firefighters should not be paying more in taxes than billionaires. That’s why I support a billionaire minimum tax and will ensure corporations pay their fair share.

Looking at trends in the richest people in the world, in 10 years the richest person should hit a trillion USD. That’s insane. No one should have that kind of wealth. There’s something very wrong with that picture… and most of the wealthiest are from the United States.

I’m confident open source models will overtake closed ones at some point and be democratized. I haven’t compared llama 3.1 with gpt-4o, but it’s still pretty darn good 😊