• Google Maps has correctly marked Lake America and the crazies are being crazy.  (Daily Beast)
  
  Oh no.
  
  
  
• Speaking of crazies, they're upset that America's education system has stalled out 99% control by communist infiltrators.  (LinkedIn)
  
  Anyway.
  
  
  

• Nvidia's DLSS 5 upscaling technology is here.  Don't use it.  (WCCFTech)
  
  Play Deep Rock Galactic - a fun but not graphically challenging game - on an RTX 3080 - an older but still powerful card - drops frame rates from 138 fps to 4.
  
  
  
• Omarchy Linux, the new flavour, has a teeny tiny bug.  (0xcc)
  
  The default user has access to everything.
  
  
  
• Chess.com has launched a poker website.  (The Verge)
  
  Apparently their Eight Queens Hold'em site wasn't bringing in the views.
  
  
  
• Oops.  My language blew its budget; compiler, runtime, and standard libraries now come to 7.5MB or 2.5MB compressed, making it 8% the size of Python.  The two culprits were TLS (a.k.a SSL, the encryption used by every website), and regular expressions across Unicode, each contributing 20% of the total.  For embedded use it automatically drops those and clocks in at a far smaller size.
  
  Also landed today: A full "math" library (those funny buttons on a scientific calculator), alongside autovectorising vector and matrix functions (vectors 100x faster and matrices 1000x faster than native), and a more general linear algebra library built on those, hashing, cryptography, passwords, HTTP client and server modules, and a curl / Python Requests style library.
  
  Claude has actually impressed me; it's a brand new programming language with precisely zero training data, and it's writing working code in it.  Yes, it has specs; yes, it has access to the compiler and runtime internals; but the former is up to the developer and the latter will be open source.
  
  I pitted it against the Programming Language Shootout - since renamed - and it came out four times slower than Ruby.  Not great, but that's the first time I've looked at performance numbers, and a quick code review found the problem and two, possibly three different fixes.
  
  Update: Tried out two of the fixes.  One improved performance by 30%.  The other improved performance by up to 56,000%.  I'll do both, but the second trick is clearly the one to chase first.
  
  Further Update: Definitely needs both.  That 560x speedup was on one specific benchmark; others are much more fussy.

Posted by: PatBuckman at Monday, August 31 2026 10:51 PM (6eVrQ)

Hide Comments | Add Comment