Is There a Future for Decision Transformers?

The decision transformer really combines all these things in a way that maybe is before they've been more separate. So hopefully you can build a really good video model and then use that as a foundation model to like fine tune by adding rewards so that it kind of like builds the entire cake in one large giant transformer I think that would be awesome. There's no particular work that I'm able to highlight as saying oh yeah this is an incredibly awesome follow up. My feeling is people should just try to make a much larger model with a lot more trajectories and train it and that would be the real deal.