Stockpyl v1.0.2 is out! A few new EOQ-related functions, a bug fix or two, and a change in license from GPLv3 to MIT.

pypi.org/project/stoc...

If you're looking for a break from the news, try my crossword puzzle in today's NYT! It was a lot of fun to construct and I hope you find it fun (and pretty easy) to solve! #nytxw www.nytimes.com/crosswords/g...

I wrote today’s @usatoday.com crossword puzzle. It’s free to play (today) and on the easy side. Enjoy! puzzles.usatoday.com

Lately I've been thinking a lot about this visualization of wealth inequality. (@kanarinka.bsky.social and @laurenfklein.bsky.social might call it a "visceralization".) It's a few years old but is still so effective. eattherichtextformat.github.io/1-pixel-weal...

Excited to discuss our Lehigh U Algorithms and Social Justice course w/Prof Suzanne Edwards at Lafayette Coll this Weds. And humbled to help kick off a series of talks on AI for inclusion, diversity, equity, and access, with later speakers like @poetofcode.bsky.social and @alexhanna.bsky.social.

Congrats to former #Lehigh #ISE PhD grad Zümbül Atan on becoming the new chair of the Operations, Planning, Accounting and Control (OPAC) research group at Eindhoven U of Technology! www.tue.nl/en/news-and-...

There's a way to tell ChatGPT not to rewrite your prompt, so we tried it, and sure enough, here's the image:

In other words, DALL-E isn't actually less biased than it was, ChatGPT just gives it explicit instructions to avoid *appearing* biased.

Then @jeremymack.bsky.social told me that ChatGPT rewrites prompts before sending them to DALL-E. He prompted "generate an image of four doctors in a lab." Here's how ChatGPT revised the prompt and the resulting image. (To the naked eye, this appears to be 6 drs, not 4, but I digress.)

I said at a workshop at Lehigh today that a few yrs ago, a ChatGPT/DALL-E prompt like "generate an image of 4 doctors" would give an image of 4 white male drs, whereas now the drs would be more diverse--but that's only bc of superficial guardrails--the underlying data and model are still biased.