RT @ego_virgo: The @LIGO - Virgo - @KAGRA_PR Collaboration is happy to announce the detection of the merger of the most massive black holes….

Minst bekende en ook mooiste straatje van Valkenburg @magnifiekZL

Whoa it is bye bye cursor. Claude code for the win! No need for a GUI anymore, since i can do everything from my VPS/iPad command line.

Misschien wat benzine inslaan. Je weet maar nooit. #wcroltweet.

Jezus, alsof het mij wat boeit of het een bekende boer is @delimburger 50.000 beestjes ter ziele. Pas die kop es aan

Owh man, Claude advised me to put an .env into a public html folder. Wow. And that was not the first security issue. Also adviced to put a git repository inside a public html. Happy vibe coding!.

Pretty underwhelming so far. #WWDC25.

Amazing tech in Maastricht. Visited the ET Pathfinder site today.

#claude4 is here!

Hoog bezoek bij de Curfsgroeve vanochtend @magnifiekZL

Als letterlijk de wereld in chaos verzandt dan is er altijd nog L1 met het verheugende nieuws dat de goudvis van Andre Rieu door de brandweer is gered. Waarvoor dank.

Meta claims Llama-4-Scout can handle 10M tokens with chunked attention. Tried it with 90.3M chars split into 452 chunks. Getting weird patterns (because of base model?) and 71GB memory usage. Not sure if this is the right approach or if the claim holds up. Anyone else tried

Wow! Quite a performer. #llama4 scout on MBP M3 max 128 (mlx) . Prompt: 76 tokens, 100.317 tokens-per-sec.Generation: 2000 tokens, 31.601 tokens-per-sec.Peak memory: 61.114 GB

Ah of course LMstudio is not supporting it. That is understandable. Switch to MLX chat or similar to do some inference testing, i am especially interested in maximizing the context window and performance on Max m3 max 128 vs my current RAG setup (with postgresql).

Now testing #llama4 scout with LMstudio using this one: The one showing up in search is approx 200GB and not correct. This one has a correct size.