some quick thoughts on rsi 1) i think that it is important to separate two different concepts of rsi: (a) fully automated model development and (b) accelerated development of model capabilities 2) it is important to separate these because they each have a different effect on the shape of the future; 3) fully automated model development would mean that frontier labs would look more like oil fields than quant shops; you would not need as much human expertise to operate them 4) this would make them much easier for to nationalize; it would be much harder for employees to revolt against management or government action they don't like 5) it could also decrease the rate of algorithmic diffusion between labs (no employees to defect) and potentially between countries (for the same reason) 6) the second concept is an acceleration of capabilities; this could come from automated model development or from something else (e.g. prev was reasoning models) 7) acceleration of capabilities is important because it could lead to faster automation of office work and transformative technologies (e.g. bio, materials, etc...) 8) this is very relevant to future employment patterns, lab revenues, when and how the government will react to ai, geopolitical and strategic effects of ai, etc... 9) it is not clear to me that fully automated model development necessarily means that you will have and accelerated development of model capabilities 10) fully automated model development could just be priced into the eci chart; people have always been using old tools to develop new tools, etc... 11) we don't know what the relationship is between fully automated model development and acceleration; i would be interested to read forecasts here