Already it’s impossible to control the flow of slop in a team of non-negligible size. The natural response is to fork and protect anything on your critical code paths.
But what’s the end state? Everyone has their own stack from OS up and we interact via IETF style RFCs?
first pass at dlpack support in tinygrad: github.com/tinygrad/tinygrad…
this should unlock support for nccl4py. excited to play around with that later as well!
finished a fun end of year project: implementing FFT and LoRA in tinygrad: dxuuu.xyz/peft.html
Found a nice bug in tinygrad’s llama 3 1b implementation: github.com/tinygrad/tinygrad…
Daniel Xu retweeted
Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is “Defeating Nondeterminism in LLM Inference”
We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to prompt engineering. Here we share what we are working on and connect with the research community frequently and openly.
The name Connectionism is a throwback to an earlier era of AI; it was the name of the subfield in the 1980s that studied neural networks and their similarity to biological brains.
thinkingmachines.ai/blog/def…
One thought I’ve been coming back to is that code review skills are going to be increasingly more valuable as AI generates more and more code.
Even in the span of a few months Claude code is incredibly more capable. I mostly just sit back and review code now.
