[matltc]

I did read it a while back. Was curious what parent was referring to specifically

[9991]

March 2027 -> Neuralese recurrence and memory

> For example, perhaps models will be trained to think in artificial languages that are more efficient than natural language but difficult for humans to interpret.