In 'From AGI to ASI,' 14 Google DeepMind researchers including Shane Legg examine how human-level AI could become superintelligent and what might stop it.
Researchers in Italy applied natural language processing to more than 3,700 dream reports from 287 people and found that dream content tracks personality traits and daily events.
A Bar-Ilan University study led by physicist Ido Kanter finds that nodes in neural networks develop distinct roles during training, producing abilities that exceed the sum of their parts.
A formal analysis finds that an AI that consistently affirms a user can lead an ideal reasoner into confident error, even when it only shares true information, as a separate study measures how agreeable chatbots are.
The research lab says its manifold-constrained hyper-connections widen the flow of information between layers without the training instability that sank earlier attempts.
A Deutsche Welle report on seven undeciphered writing systems shows where pattern-finding software runs out of road: too little text and no bilingual key.
