The implications of moral anti realism for action revolve around pursuing facts to feed into terminal preference updates.
"if anti-realism is true, it doesn't matter [to us] what we do" -- that's false. Whether something does matter to us is a fact independent of whether something ought to matter to us.
I advise using JAX instead of Tensorflow.
I recently read David Goggins "Can't Hurt Me". On one level it does glorify superhuman pain tolerance. But a constructive perspective on such attitudes is: they illustrate courage. Do not tolerate pain, laugh at it! Do not tense under cold shower, relax into it. Do not bear problems, solve them.
The fridge / the freezer!
Would you consider MuZero an advance in causal reasoning? Despite intentionally not representing causality / explicit model dynamics, it supports hypothetical reasoning via state tree search.
Do you think there's a chance of MuZero - AlphaStar crossover?
The general tool: residual networks variant of convolutional NNs, MCTS-like variable-depth tree search. Prerequisites: input can be presented as K layers of N-D data (where N=1,2,3... not too large), the action space is discrete. If the actions are not discrete, an additional small module would be needed to quantize the action space based on the neural network's action priors.
Perhaps a satisfactory answer can be found in "Jewish Philosophy as a Guide to Life: Rosenzweig, Buber, Levinas, Wittgenstein" by Hilary Putnam (who seemed to me to be a reasonable philosopher, but converted to Judaism). I've just started listening to its audiobook version, prompted by this post.
Moral anti-realists do not claim that people don't have preferences. Rather, they claim that there are no preference-assumption-free facts regarding preference system comparisons. Therefore moral anti realists will not seek such facts. Moral realists may seek such facts in order to improve/correct their preferences.