@EitanTurok@__tinygrad__ I don’t need to have it in a single shot.
You can run it multiple times, choose the best one, and then run another swarm of agents with the task of making it tiny grad style.
They’re not humans, you need a different workflow with more brute force.
@cotyledonlab@levelsio A cli tool or even some APIs can do the same thing without a new protocol that simply redefine what we already have and understand.
@JohnStrongHodl@levelsio Frameworks are abused, code generation further reduces their utility.
AI simply exposes a mistake that we know existed, but we ignored because the reward for solving it was too little. Until now.
@garrytan It think it will happen before that point. It may simply happen because the overhead of increased headcount slowdown the existing engineers more than the production of a new engineer.
@PaulSolt I will definitely try it when it is available.
Last week I opened Xcode again after over 5 years, and I’m experimenting with different skills/MCPs and workflows to test some ideas.
Claude&co made app development fun again for me 🙂
@xsteenbrugge@OdinLovis It’s a distillation, you use a bigger model to train a smaller model so you can have almost the same performance with less computation.
Z-image can run for “free” on your laptop, Banana requires google data centers.
@BenJamminAsh@EHuanglu Stories also have personality, animated movies already exist and billions of people watch them.
Viewers want stories, and beautiful stories can be told already without actors.
@GariB4lda@gecgreek Uno che deve mescolarsi in una folla di repubblicani per fare un omicidio e poi scappare, per esempio.
Andarci con la kefiah e la maglia arcobaleno avrebbe dato un filo nell’occhio. Ma un filo solo.
@rameerez Because they have to shrink the Apple Vision into a pair of glasses and build millions of them before their competitors.
You need intermediate steps, and you need to test them at scale.
It’s a smart move, and they might also recover some of the investment by selling some AIRs.
@CharlyMatt Ho letto i “perché” prima di commentare, e sono sinceramente spiazzato perché rendono chiaro che non è un errore dato da inesperienza, ma un metodo che deriva da “10 anni di esperienza”.
Abbiamo un concetto diverso di cosa vuol dire essere professionali chiaramente.
@theovonscousin@esrtweet@StefanMolyneux Yes, after all data is available, I just “know” the result.
I did not listen to the debate in my brain, but I can recall it as a dialogue if I want.
(But I rarely do it, I just start to feed the data for the next step. )
@realPolemikoh La differenza scompare ma non il problema qualitativo: qualitativo: chi arriva ingrossa le file dei marginalizzati e di conseguenza i reati.
Per ridurre reati o migliori i processi di immigrazione o riduci i flussi. La prima soluzione è migliore, ma molto complicata.