From start to finish, the #LK99 saga showed us one thing: the West is massively behind China in basic scientific curiosity and agility. Our labs' top priority is writing yearly reports to secure next-term budgets. The actual science can always wait.

Gotta love this level of intelectual honesty!

Here's a cute little gift of the pandemic: public's mistrust of science escalated to a level where politicians can rack up support with claims that the threat to human prosperity isn't global warming, but the transition to clean energy. x.com/VivekGRamaswam…

History of excuses for not doing your job: 1. lights are out 2. internet is down 3. StackOverflow is down Right now, ChatGPT is down, and I haven't even checked StackOverflow. My day is done.

Just got across this classic, it was fun to rewatch! It reminded me how much the widespread understanding of AGI improved in the meantime. You won't find many disputes along these lines today. youtube.com/watch?v=hEUO6p…

I've been hacking away in the shadows for #1BRC, got encouraged to submit when I saw the lively back-and-forth that goes on among the contestants on their PRs. Looking forward to some nice convos! github.com/gunnarmorling/…

For fun and curiosity, I implemented the same #1brc solution in Rust, and in knocks Java out of the park: 5.3 seconds on the same Hetzner CCX33 instance as used for the leaderboard. Just 0.15 branch misses, and 0.02 cache misses per input row!

My unofficial #1brc Full Dataset Leaderboard: 1. thomaswue 9.0 2. mtopolnik 9.1 3. royvanrijn 14.3 4. hundredwatt 35.7 5. merykitty 71.7 Honorable mention: artsiomkorzun 7.2, but invalid output. Probably an easy fix away from the podium. The second battle is heating up! NB:

Kind of unexpected, but my little piece of Rust still scores #1 on the 10k #1brc dataset! It's a straight-up solution, all Safe Rust except mmap (which is fundamentally unsafe regardless of language). github.com/mtopolnik/rust…

Interesting bit on #1brc: merrykittyunsafe is the best implementation using the Vector API, and it sits at the throne currently with 2.575. Yet, on the 10k leaderboard, its performance tanks. Why? The relevant difference is the statistical distribution of the station name

Remember the Monty Hall Problem? Prompted by this letter, I watched a dozen episodes of Let's Make a Deal. Not a single one of them sported "The Monty Hall Problem"! Contestants were never offered to switch doors.

The Billion Row Challenge (#1BRC) - Step-by-step from 71s to 1.7s questdb.io/blog/billion-r…

#1BRC contestants and those who followed it probably know what I mean by "merykitty's magic SWAR". I sat down and tried to explain those awesome 8 lines of code in ~3,000 words of English. Comes with the seal of approval from the man, Mai Đặng Quân Anh ! questdb.io/blog/1brc-mery…

Does "vpmovzxbd" Scare You? Here's Why it Doesn't Have To! questdb.io/blog/does-vpmo… Another wonderful blog by Marko Topolnik 🤓

Today in QuestDB pt2: Marko Topolnik made length(varchar) SQL function faster with the following SWAR code. Previously it was using a much slower UTF-8 parsing routine. This reminds me of cool tricks from Daniel Lemire blog.

I don't think this ever happened when I was using GPT: I present my coding problem and the route I'm taking, then ask for a specific detail along the route. Claude gets what I'm doing, tells me that's too complex, and here's a better way. BTW the first attempt didn't work,

"A survey on the evolution of stream processing systems" Banger of a paper by Marios Fragkoulis, discussing ordering semantics, fault tolerance, scalability, state management and more, of different stream processors. 👉 link.springer.com/article/10.100…

Combine Java and Rust Code Coverage in a Polyglot Project: questdb.io/blog/rust-cove… A guide by Marko Topolnik himself!