Links for 2023-05-16
MEGABYTE: Predicting Million-byte Sequences with Multiscale Transformers https://arxiv.org/abs/2305.07185
A new European AI regulation proposal would make any "American opensource developer" that hosts an "unlicensed LLMs" on GitHub & available in Europe liable for "€20,000,000 or 4% of worldwide revenue" https://technomancers.ai/eu-ai-act-to-target-us-open-source-software/ (Full PDF from the European Parliament: https://www.europarl.europa.eu/meetdocs/2014_2019/plmrep/COMMITTEES/CJ40/DV/2023/05-11/ConsolidatedCA_IMCOLIBE_AI_ACT_EN.pdf)
“We gave GPT 464 moral scenarios from past papers, and asked it to make moral judgments--they correlated .95 with human ratings.” https://www.dropbox.com/s/99wq0yyk6n85bu3/GPT%20Morality.pdf?dl=0
Study: AI models fail to reproduce human judgements about rule violations https://news.mit.edu/2023/study-ai-models-harsher-judgements-0510
GPT-4’s Maze Navigation: A Deep Dive into ReAct Agent and LLM’s Thoughts https://ekzhu.medium.com/gpt-4s-maze-navigation-a-deep-dive-into-react-agent-and-llm-s-thoughts-b1823fb266ee
Gameface: A hands-free, AI-powered gaming mouse https://blog.google/technology/ai/google-project-gameface/
What Midjourney thinks professors look like, based on their department https://www.reddit.com/r/midjourney/comments/131ebyk/what_midjourney_thinks_professors_look_like_based/
First UK baby with DNA from three people born after new IVF procedure https://www.theguardian.com/science/2023/may/09/first-uk-baby-with-dna-from-three-people-born-after-new-ivf-procedure
On a related note, from a layman's perspective, is there a more misleading concept than heritability?
When laypeople talk about genetics, they are most interested in genetic determinism, i.e. whether traits are genetically determined, and their fitness. But most genetic studies talk about heritability: "the proportion of variation in a given trait within a population that is not explained by the environment or chance".
This is confusing because natural selection tends to remove the genetic variance of traits that are correlated with fitness, reducing the heritability of the trait.
A common example to illustrate this is the number of fingers a person has and whether they put on lipstick. The former is mostly genetically determined, while the latter is mostly not. However, the number of fingers we have has very little heritability, as most of the variance is explained by environmental influences such as accidents. Lipstick application, on the other hand, is highly heritable in many populations because it correlates almost perfectly with the sex chromosomes.