Last week I published a 30-minute documentary about SB-1047 I had been working on since September.
Here's what I learned.
In an early version of the bill, as effective compute becomes cheaper, the bill would be covering more and more models, not just the frontier[1]. This is what led to June's amendment, introducing a $100M training run threshold.
When following how some of the talking points have been repeated in multiple letters against the bill, it appears that the initial lobbying from Andreessen Horowitz was repeated by Fei-Fei Li, then members of Congress, and eventually culminated in Nancy Pelosi's opposition statement
Mostly because of algorithmic improvements, but also simply because of compute getting cheaper. In the doc the audio simply says "compute getting cheaper" not "effective compute" because we wanted to simplify it for a broad audience.
Watch the full documentary here: https://youtu.be/JQ8zhrsLxhI