Tim Dettmers
A researcher who works on making large AI models run on small hardware, best known for the bitsandbytes quantization library.
Tim Dettmers is an academic, at Carnegie Mellon and the Allen Institute for AI, whose work is mostly about one question: how do you fit a very large model onto hardware that normal people own? He created bitsandbytes, a library that shrinks models by storing their numbers with fewer bits, and which sits quietly underneath a great deal of the local AI software people use without realising it.
He also writes, at length and in plain language, about GPUs, research careers and where he thinks the field is going. His hardware guides have been the standard reference for people choosing a graphics card for machine learning for years. He argues that small university labs, not just well funded frontier labs, will produce much of the important work ahead.