A few days ago, I came across the paper “Demoing Stochastic Parrot: A Candid AI Cohabitant” by Chang et al. This paper—published by MIT researchers at the ACM Conference on Human Factors in Computing Systems in April of this year—presents its authors’ work on a physically embodied “AI cohabitant” that they call a “stochastic parrot”. It begins with the following passage:
This post is part 3 in my ongoing series on fast Fourier transform algorithms. You may wish to read part 1 and part 2 before reading this article.
This post is part 2 in my ongoing series on fast Fourier transform algorithms. You may wish to read part 1 before reading this article.
Confucius once said:
I’m planning to write a series of posts about fast Fourier transform algorithms. This first post covers the Cooley-Tukey algorithm, which is the original and most well-known FFT algorithm.
U.S. Congressional Representative Paul Gosar of Arizona1 recently reintroduced proposed legislation to ban the Optional Practical Training program (OPT).2 OPT is a program that allows international students attending college & grad school to legally remain in the United States and work in their field of study for 1 year after completing their degrees.3 Students majoring in an approved science,…
After realizing I was mildly colorblind, I made this program to generate colorblindness tests
I just read Nate Silver’s blog post, where he writes that pollsters are systematically altering their data to roughly match the average of existing polls. According to Silver, rather than releasing their findings as-is, they’re worried they’ll look uniquely wrong, and so they’re settling for blending in with the crowd. He infers this bias from the numbers that the pollsters themselves report; the…
The other day, I was attempting to develop a Knative service and try it out on my local development set-up, which was a Minikube cluster. I assumed (incorrectly) that I could build a Docker image on the host machine and it would be automatically available to Minikube. However, this is not true, because Minikube has its own Docker daemon, inside of its own virtual machine (which, if your set-up is…
TL;DR: if you are using scikit-learn 1.3.X and use f1_score() or classification_report() with the argument zero_division=1.0 or zero_division=np.nan1, then there’s a chance that the output of that function is wrong (possibly by any amount up to 100%, depending on the number of classes in your dataset). E.g. for zero_division=1.0: In this post, np.nan refers to numpy.nan