Maybe, every night, you send the AI off to "sleep" where it uses those in cache "memories" to influence the long term weights [1].
replyContext self-distillation does exist, but as is, it's used mostly in training rather than as a part of a continuous learning mechanism.
reply