Links for 2022-06-10
Metaculus: Will more than two of Gary Marcus' proposed AI achievements come to pass before 2030? https://www.metaculus.com/questions/11199/gary-marcus-agi-bet-2030/
AI Could Defeat All Of Us Combined https://www.lesswrong.com/posts/oBBzqkZwkxDvsKBGB/ai-could-defeat-all-of-us-combined
"SayCan: Do As I Can, Not As I Say: Grounding Language in Robotic Affordances", Ahn et al 2022 (language models powering robots) https://say-can.github.io/
Ion Channel Density Surprisingly Different for Human Neurons [published in 2021] https://www.genengnews.com/neuroscience/ion-channel-density-surprisingly-different-for-human-neurons/
What’s the endgame of neuroAI? https://xcorr.net/2022/05/18/whats-the-endgame-of-neuroai/
Irritable bowel syndrome: How silence and stigma around IBS symptoms is harming patients https://inews.co.uk/news/health/irritable-bowel-syndrome-how-silence-and-stigma-around-ibs-symptoms-is-harming-patients-1662167
Two meta-analyses find that "sexualization in video games was neither related to well-being/body dissatisfaction nor sexism/misogyny. Better designed studies... tended to find less evidence for effects." https://sciencedirect.com/science/article/pii/S0747563222001637
How can you increase crop yields by as much as 25%? Cut emissions of a common air pollutant. https://fse.fsi.stanford.edu/news/less-air-pollution-leads-higher-crop-yields-stanford-led-study-shows
PLA researchers say drones will use AI and publicly available photos from social media to identify, track or kill high ranking officials behind enemy lines https://www.scmp.com/news/china/science/article/3180220/china-spurred-russias-setbacks-ukraine-develop-smarter-military
Dying your hair unnatural colours is correlated with depression. So is vegetarianism. https://pjp.psychreg.org/wp-content/uploads/2022/06/3-edward-dutton-blue_hair_blues_26-35.pdf
Google just released a 442-author paper about a monster new test suite for evaluating Large Language Models (GPT-3 and the like), and in particular, their study of the language models' performance on their test suite as the number of parameters is scaled. As a striking example, see below for a neural net's ability to guess a movie from emojis (rather than, say, outputting random nonsense) as the number of parameters is gradually scaled from 2 million all the way to 128 billion.
https://github.com/google/BIG-bench/blob/main/docs/paper/BIG-bench.pdf
This should be worrying because it makes it difficult to predict future progress. They are not just getting predictably better with more training, data, and parameters but sometimes capabilities emerge in a jumpy and unpredictable way. There can be sudden phase transitions.
The left: Why has the right's confidence in the scientific community declined?
The executive editor of The Lancet:
Equity is one of the most destructive ideas ever conceived of.
The Khmer Rouge is what happens when you take the concept of 'equity' to its logical conclusion.
Even in a society with advanced genetic engineering in which everyone can be a good-looking and healthy genius won't be enough to achieve equity.
Due to sheer luck, people will eventually accumulate more wealth, connections, and thereby opportunities, than others. Outcomes will then start to diverge.
The only way to guarantee equity is by force; by continuously robbing people of wealth and status and redistributing it among strangers.