Reinforcement Learning Progress
blog.samaltman.com> ...and a really good simulated environment that captures the problem you’re solving.
The original "data is the new oil" quote was pointing out that raw data, like raw oil, requires lots of processing/refinement before the raw resource becomes something with a lot of economic value and potential [1].
In that sense, simulated environments are the oil of deep RL.
Deep RL has a lot of promise (and obv is already delivering on that promise). But when it comes to the need for high-fidelity and accurate models, we're out of the frying pan and into the fire.
[1] https://medium.com/@TalPerry/on-labeled-data-85fbaf1bdf89