www.newyorker.com

was written in a period when NLP practitioners are producing bigger (# of parameters; size of training data) language models (LMs), and pushing the top scores on benchmarks. The paper itself was controversial because it led to Gebru being fired from Google, following disagreements with her managers on conditions (withdraw, or remove Google-affiliated authors) for publishing the paper. A lot changed since mid-2021, when I initially wrote this page....

Harden, a liberal behavior geneticist, is publishing a book, “The Genetic Lottery: Why DNA Matters for Social Equality”. Watch this space as it casts doubt on prevailing social justice thesis that environmental factors, and not genetics, influence behavior or social outcomes. Check back in Sept 2022 (one year from now). Faster Human Trials In 1986, AZT (AIDS treatment) showed 1/145 deaths compared to the placebo’s 16/137. 6,000 AIDS patients were offered AZT, helping it get to the public much faster....

Bernoulli Process A Bernoulli Process is a sequence of independent \({0, 1}\) - valued random variables \(X_1, X_2, X_3, …\), e.g. \(0, 0, 1, 0, 1, 1\) A Bernoulli Process does not mandate that the probability distributions of the \(X_i\) be identical. That is up to the model that we choose. For instance, the Binomial Random Variable assumes \(\mathbb{P}\{X_i = 1\} = p \ \ \forall i\) Suppose you flip a coin repeatedly, and record \(0\) for tails and \(1\) for heads....

Fall 1990. Simons had spent 12 years searching for a successful investing formula. In 1989, his net returns were -4.0%. In 1988, he had netted 9.0%. In 1990, he netted 55.0% (77.8% before fees). With Elywn Berlekamp, game theorist professor at UCB, Simons built a model for selecting idea trades based on data. Berlekamp knew their approach was alchemic - no cash flows, interest rate forecasts, product research. Some of the trade recommendations didn’t make sense....