Hi, I’m Dustin.
I work on how language models remember things. Continual learning, long-context memory, and making it all cheap enough to run on GPUs people actually own.
I’ve been doing this computer thing for about twenty years, starting with racking supercomputer nodes for national labs and working my way up through ops, platforms, and eventually AI. These days I train models and build agents on my own small GPU fleet, and write up what happened here, including the parts that didn’t work.
Recent writing
-
Rebooting This Site
The old dustinmorgan.org was a three.js demo with a bio I wrote in one night. This is the replacement.
-
Training the Harness Instead of the Model
I fine-tuned dozens of models to make Agent Neo better at incidents, and none of them beat a better harness. So I started training the harness. It went from 5 to 10 out of 12 hard investigations.
-
Coming Soon: CIP and CTP
Two things I've been building in private: running big open-weight models across the machines you already have, and training on them too.
-
What I Got Wrong About SQ4
I published a 4-bit quantization format and called its quality indistinguishable from FP16. Once I measured it properly, standard 4-bit formats beat it. I've retired it.
-
SQ4 Is Live: Whitepaper + llama.cpp PR
We published the SQ4 whitepaper and opened a PR to bring percentile-calibrated 4-bit quantization to llama.cpp.