Links for 2022-06-03
Least-to-Most Prompting Enables Complex Reasoning in Large Language Models — “This “least-to-most prompting” method solves 99.7% of SCAN benchmark while other prompting methods solve ~16%.” https://arxiv.org/abs/2205.10625
Towards artificial general intelligence via a multimodal foundation model: "With more sensory modalities exploited for multimodal pre-training and further exploration on more advancing foundation models, we believe that we are approaching AGI" https://www.nature.com/articles/s41467-022-30761-2
Testing heritability of moral foundations: Common pathway models support strong heritability for the five moral foundations https://journals.sagepub.com/doi/abs/10.1177/08902070221103957
How BIG is the universe? Sounds like a simple enough question, but in reality, there are four different ways to answer it, and they're arguably all correct! https://www.youtube.com/watch?v=ea3e52Uu4aQ
“I believe we have not yet internalized how rapidly a middle income country can fall from grace and into utter chaos.” https://marginalrevolution.com/marginalrevolution/2022/05/the-fall-of-sri-lanka.html
Advances in power beaming: A century later, Nikola Tesla’s dream comes true https://spectrum.ieee.org/power-beaming
Higher SES people have more specialization between hemispheres of the brain: "hemispheric specialization may have evolved in human primates in a way that reveals crucial links to socioeconomic status." https://academic.oup.com/cercorcomms/advance-article/doi/10.1093/texcom/tgac020/6589935?login=false
“Life based on exotic matter could start space travel before reaching general intelligence and it will colonise a large part of the universe without creating visible transformative changes.” https://www.lesswrong.com/posts/PLzantdwa8XJMLKef/grabby-animals-observation-selection-effects-favor-the
Will AI Lead to New Creative Frontiers, or Take the Pleasure Out of Music? https://pitchfork.com/features/article/ai-music-experimentation-or-automation/
A Brutal Examination: Russian Military Capability in Light of the Ukraine War https://www.tandfonline.com/doi/full/10.1080/00396338.2022.2078044
Fredkin’s Paradox highlights something that always bothered me about Kahneman's modes of thought dichotomy:
“[[I]f I want to optimize, I must also determine the effort it will take me to optimize; however, the subtask of determining this effort will itself take effort, and so forth into the tangle that self-referential activities create.”
For all practical purposes, there really only is "System 1". Unconscious, automatic processes decide when to reason more slowly and when to stop doing so. In fact, a famous procedure used to make difficult decisions highlights how conscious, logical, and calculated reasoning is often just a tool to rationalize what we subconsciously want:
"Hold the coin in your hand for a minute, visualizing the two options on either side. Take a deep breath, and toss the coin into the air. Suddenly it doesn't matter which side the coin lands on – because you know which side you're hoping it lands on. You have your answer."
This is also why the biggest advantage of the profoundly gifted is not their complex logical reasoning skills but their superior intuition. They can intuitively locate the golden needle in an infinite haystack and terminate thoughts that are intractable by predicting which thoughts will quickly lead to infinite recursion or combinatorial explosion. In other words, it's system 1 rather than system 2 that allows them to make useful conjectures.
The same is true for all kinds of successful people. They possess superior noise filtering. Their gut feelings are able to correctly dismiss certain information.
System 1 is the very foundation of bounded rationality (rational decision-making under limited resources). It is the reason why we don't get mugged by game-theoretic agents threatening us with their magic powers from outside the Matrix: our gut feelings call bullshit.
Evolution equipped us with numerous defenses against various sorts of memetic hazards that would cause our minds to crash: heuristics that terminate infinite recursions and intractable problems. For example, “near bias” allows us to avoid indecision and unbearable suffering by both ignoring long-term consequences and by focusing on people who are close to us.