Eric Jang – Building AlphaGo from scratch

AI assistants helped researcher Eric Jang run experiments while rebuilding a Go-playing system, but choosing the research direction remained harder. He describes assistants that could change training settings, rewrite experimental code and produce plots from a requested question. This made their search for improvements more flexible than trying a predefined list of settings.

Jang also says he had to recognize unproductive research paths and prompt investigations into infrastructure bugs. His experience separates two kinds of scientific work: carrying out an experiment and deciding which experiment would resolve the important uncertainty. Automating the first can be useful while the second still demands substantial researcher involvement.

“Thanks to LLM coding, what took a whole team of research scientists at DeepMind and millions of dollars of research and compute can now be done for a few thousand dollars of rented compute.”

Eric Jang

“In deep learning, initialization is everything. You always want to initialize your research project to something as close to success as possible”

Eric Jang

“This was a breakthrough that I think most people don’t even fully comprehend today, how profound that accomplishment is.”

Eric Jang

“Because Go is such a combinatorially complex game, you cannot afford to build the tree in advance and then search it. You must search while building the tree.”

Eric Jang

“The compute required to be the first to do something is always much larger than the compute it takes to catch up.”

Eric Jang