Links for 2022-05-24
Imagen: Google's new image-generator (even better than DALL-E 2?) https://gweb-research-imagen.appspot.com/
Visually-Augmented Language Modeling: Outperforms the text-only baseline with substantial gains of +8.66% and +37.81% accuracy on object color and size reasoning, respectively, by retrieving images relevant to the context. https://arxiv.org/abs/2205.10178
These nine cognitive psychology findings all passed a stringent test of their replicability [published in 2017] https://digest.bps.org.uk/2017/06/05/these-nine-cognitive-psychology-findings-all-passed-a-stringent-test-of-their-replicability/
New experimental evidence from Liberia shows cash payments + cognitive behavioral therapy can reduce antisocial behaviors (such as selling drugs or committing robberies) by about half https://www.nber.org/papers/w30049
Are You Smarter Than a Random Expert? The Robust Aggregation of Substitutable Signals https://threadreaderapp.com/thread/1457781670430404616.html
Would you pay 5 cents for 5 IQ points? https://ideassleepfuriously.substack.com/p/would-you-pay-5-cents-for-5-iq-points
Innovation is not linear. Not only does basic science feed into applications, but applied science feeds back into basic science. https://www.worksinprogress.co/issue/innovation-is-not-linear/
“What's the difference between parasites and predators? I'll re-ask this question in different forms as we go. I invite you to figure out the answers for yourself.” https://tsvibt.blogspot.com/2022/05/the-lion-and-worm.html
New Stanford Encyclopedia of Philosophy article, "Normative Theories of Rational Choice: Rivals to Expected Utility" https://plato.stanford.edu/entries/rationality-normative-nonutility/
What is play? Researchers showed kids pictures, asking, "Does this show play or NOT play? Usually, if there was an adult in the pic, even if the kids were smiling, they said: That's not play. https://www.psychologytoday.com/us/blog/freedom-learn/202205/what-is-play-how-children-define-it
Popular education in Sweden: much more than you wanted to know https://www.lesswrong.com/posts/tjxgbovwc5Ft7wrtc/popular-education-in-sweden-much-more-than-you-wanted-to-1
There are various dimensions of understanding.
If you understand <X> then...
1. you possess compressed data about X in the form of a simple model, theory, or axiomatic system.
2. you can use #1 in order to make predictions or derive new facts about X.
3. you can use #1 in order to solve a concrete problem involving X.
4. you can derive general rules and concepts from #1 and apply them to different but related problems.
5. you can communicate knowledge about X by teaching/explaining/programming something to perform abilities #1-4.
6. you are able to rederive or rediscover abilities #1-5 from first principles.
7. you are able to consciously execute all of the aforementioned abilities.
P.S. I once reread a proof and noticed that I misunderstood a critical step. This shows how easily one can end up with a fake sense of understanding and how important feedback loops in the form of exercises are.