Cool, $200 for a model that can't remotely match $20 Claude Sonnet. This will be a huge hit I guess.
Posts
Yesterday I was in a call with my daughter's teachers about an exchange project. They reported they are unable to do exchanges with UK because families refuse to send children here in Sicily :( This doesn't happen with France, for instance. They don't understand causes to fix: UK friends, hints?
For many years people said that Hacker News was ruined, full of noobs for the most part, and I kept saying: it's not that bad! Not like the first years, but it's still a cool place. Well now, in 2024: OK YOU WERE RIGHT, HAPPY NOW?
Concurrent programming is hard, but also so oddly satisfying and the design space is larger than expected. Once I'll see this program that I'm writing working very well with read/write locks, optimistic insertion and the like, I'm sure it will be a good few seconds (after which life sucks as usual).
HNSW: you write the first implementation and think, wow, that was simple, just following the paper general idea. A few weeks and iterations later: wow, I had to invent half the algorithm that was missing in the paper.
With Redis 8, there is a return on investing in the community edition (the code you just download from Github), probably what made this possible is the license change. Anyway interesting new features / improvements:
So @josefprusa just unveiled a printer that will change the short term story of the 3D printing market, storngly limiting Bambu advantage. Yet I believe that us loyal MK4 and then MK4s kit buyers (or MK4s directly) with purchase proof, should get kit at lower price.
I love this trick when traversing graphs where I can't visit already visited nodes. At every scanning: graph->epoch++, then as I visit a node, node->last_visit = graph->epoch. Much faster than a collection of visited nodes, but threading hostile... Other alternative: bitmap + node_id, but sucks.