I generally agree with the premise that RSI will probably slow down due to increasing recalcitrances, but I think focusing only on single agent model capabilities and improvement recalcitrance is not optimal, and of course indexes like ECI are only single agent. There are potential improvements possible for the ability of multi-agent swarms to organize themselves in research and thus accelerate improvement even against increasing recalcitrance on improvements to single agents. Under these conditions, we may get a preconditions different than just β<0 or... (read more)
I generally agree with the premise that RSI will probably slow down due to increasing recalcitrances, but I think focusing only on single agent model capabilities and improvement recalcitrance is not optimal, and of course indexes like ECI are only single agent. There are potential improvements possible for the ability of multi-agent swarms to organize themselves in research and thus accelerate improvement even against increasing recalcitrance on improvements to single agents. Under these conditions, we may get a preconditions different than just β<0 or... (read more)