Found this interesting paper today while searching around the keywords "attractor language networks". This part I found especially interesting:

*Attractor Models outperform existing models across two regimes, large-scale language-model pretraining and reasoning with tiny models.*