Problem with the LLM is that you need a bucket/hash list for each and every 8/16/32-sized token. And each URL are quite lengthy. And a good number of citation

Problem with the LLM is that you need a bucket/hash list for each and every 8/16/32-sized token. And each URL are quite lengthy. And a good number of citations can be found hanging off of a token.