#IlyaSutskever discusses the challenges of #AI #modelgeneralisation, comparing it to #humanlearning. He suggests that the current focus on #RLtraining, driven by evaluation metrics, might be limiting model adaptability. Sutskever proposes that expanding training environments or improving generalisation from pre-training data could enhance model performance across diverse tasks. https://www.dwarkesh.com/p/ilya-sutskever-2?eicker.news #tech #media #news
Ilya Sutskever – We're moving from the age of scaling to the age of research

“These models somehow just generalize dramatically worse than people. It's a very fundamental thing.”

Dwarkesh Podcast

Chimps & Humans: Toolmakers United - Karolina Westlund on Huberman

#animals #humanlearning #humanevolution

#FoldingAtHome got my room nice and toasty (with some help from #EisteinAtHome on another device). And though I should have been sleeping, I learned how to spot pulsars instead.

#DistributedConputing #CitizenScience #HumanLearning

People tend to underestimate the power of FAFO in human learning. I'm not just talking about "you do something you were warned was stupid and you get hurt" I'm talking about in terms of basic knowledge. Everytime you hear something that you think " who needs to be told not to do that?" Someone probably FAFOed so you don't have to. All regulations are written in blood and all are written in FAFO. #FAFO #humanlearning
@datatherapist @trochee Thank you! On a more serious note: I do use algorithms, and I am fully aware of the drawbacks for environment and society. One of the things I am working on is using #HumanLearning strategies (because we are efficient learners) and see if these can be used in algorithms to make them less data hungry. I also work on other languages than #English (#Maltese and #Kinyarwanda in particular). I know it doesn’t absolve me of all my sins. But I try.
Brain cells in a lab dish learn to play Pong — and offer a window onto intelligence

A dish of brain cells learned to play the 1970s video game Pong. The research could help computers become more intelligent

NPR
RT @gbillois
Meanwhile, in a parallel universe… #HumanLearning en avant première, LA tendance de 2075 !! #IA #AI #Deeplearning #MachineLearning #deepfake