Great read. Thanks for sharing. I have considerable expertise in recommender systems and I am interested in the intersection of recsys and LLMs. I would love to learn your thoughts on RL in recsys and LLMs or host you in our talk/GD series at Meta (Menlo Park) if you are interested.
Just like the mentioned examples that use the output of a language model as a model of human behavior, recently I've been trying to use a language model as a regularizer for an image autoencoder with discrete hidden representation. The idea is that the hidden code can look like a text sentence and the penalty from this regularization would be the negative log-probability, informed by the language model, of the produced code.
Great read. Thanks for sharing. I have considerable expertise in recommender systems and I am interested in the intersection of recsys and LLMs. I would love to learn your thoughts on RL in recsys and LLMs or host you in our talk/GD series at Meta (Menlo Park) if you are interested.
My current direction on RL in recsys
https://open.substack.com/pub/recsysml/p/stop-predicting-ctr-start-optimizing
Just like the mentioned examples that use the output of a language model as a model of human behavior, recently I've been trying to use a language model as a regularizer for an image autoencoder with discrete hidden representation. The idea is that the hidden code can look like a text sentence and the penalty from this regularization would be the negative log-probability, informed by the language model, of the produced code.