it occurs to me that quantifying the value of completed labour is extremely important now if you can accurately "position size" your token spend: identify accurately how much a task is worth, then everything else just flows downstream of that.
in allocating RL effort (synthetic and human) you already have numbers on completion success rates and token spend. the main non-obvious factor left is deciding which tasks are valuable. if u have ~the full pipeline, economically valuable RL proceeds much faster.
it occurs to me that quantifying the value of completed labour is extremely important now if you can accurately "position size" your token spend: identify accurately how much a task is worth, then everything else just flows downstream of that.
might pursue/continue more of such ideas at @hud_evals . lmk if interested ig ? https://github.com/Moneybench/moneybench
it occurs to me that quantifying the value of completed labour is extremely important now if you can accurately "position size" your token spend: identify accurately how much a task is worth, then everything else just flows downstream of that.