The document Amodei wrote to explain his views on scaling to his colleagues — titled “The Big Blob of Compute Hypothesis” — became a classic among early OpenAI researchers. It described, in florid specifics, why Amodei thought that most AI researchers were being too clever. They would engineer all kinds of complicated solutions to whatever problems their models were having, when a much simpler solution — just make the thing bigger! — would work better.
Roose goes on to describe part of what is in the latter part of the document.
Amodei also thought that AI safety researchers suffered from micromanagement. Some researchers in his orbit, like the ones at Eliezer Yudkowsky’s MIRI (more recently in the news for other reasons) were arguing for a cybersecurity-like approach to AI safety, engineering complicated safety protocols to prevent models from going off the rails. Amodei disagreed. He thought that a better way to create safe AI was to get the high-level conditions right, and let models figure out on their own what was safe or unsafe.
I was really looking forward to reading this! Unfortunately, he declined to publish the complete document.
I am publishing only the first 10, since the rest is very dense and technical and consists mainly of Amodei arguing with various other AI safety researchers about why their suggested approaches are wrong.)
This is extremely unfortunate. If anyone knows Roose, I think it would be worth seeing if he would be willing to publish the remainder.
Are you able to say more about how you trained for honesty?