Tensor Cruncher

People always ask for a list of what they should learn to get into ML systems, inference, etc. any of these lists you see for someone starting out is WRONG so let me tell you how you should be thinking. First off the reason they are wrong even if the list is genuinely accurate problems is the size of the space. In comparison with ML this is like having a set of hyper parameters that are far larger than necessary and with a search space in each that is overly large. Sure you might find the right combination eventually but you are more likely to run out of compute or hit a local minima. The real way to break into any space or to learn more is thinking in layers. You want 1. Strong fundamentals in ML if you are going to anything in the space 2. Strong programming fundamentals Ok cool so you are through layer one now what? From here we need to learn the specialization. This comes in T shaped development. You need to learn a small amount of everything you can (not too deep) and then really triple down on one or two topics. You might branch out and find interesting things you want to learn along the way and you can explore those a bit too. Find your path. So next time you see a giant scary list of topics don’t freak out and don’t really listen to all of it. Find what YOU want to focus on and don’t freak out. The only other place here is keeping up. So much new tech I would highly suggest having a small bit of familiarity. Still too much to learn but just know about what’s going on. Good luck friends. If you want advice on small sets of things to learn based on your interests let me know. Besides that don’t listen to any silly engagement bait lists

Meta burns $2.65B a year on AI tokens. at $300K for a Meta engineer, that's enough to pay ~9,000 engineers for a full year. now ask yourself: since the layoffs, has Meta shipped anything that feels like 9,000 engineers’ worth of output?

ANTHROPIC CEO WARNS OPEN-SOURCE AI IS ON A "VERY DANGEROUS PATH"

a TRILLION parameter model by a FOOD DELIVERY startup btw. maybe a dumb question, but how do these seemingly random Chinese companies have the resources/talent to make their own models?

and OpenAI's pre-training team acknowledge how the mother of all pre-training runs will eventually have to be decentralized protocols.

We recently obtained the highest-resolution 3D images of the human brain ever taken from outside the skull. This is the first look. Introducing Aleph, a research lab building brain interfaces for the telepathic future. (1/n)