OpenAI is testing out a new strategy for finding enough GPUs to keep up with demand: bumper stickers

Finally the solution to a problem we all face: it's too hard to commemorate a dumb tweet as a durable cuneiform tablet dumbcuneiform.com

The Jones Act makes shipping between American ports unreasonably expensive, especially for islands like Hawaii. We need to get rid of the Jones Act to make it cheaper to buy American.

Coreweave's business model is pretty wild:

If we're undoing all the bad decisions from the 2010s, can we get plastic straws back?

What's the logic for Instacart/Uber/Doordash constantly offering giftcards for 20% off? The discount is large enough to make delivery approx same cost as in store.

Why have TV shows gotten so ridiculously expensive to produce? I would have thought the cost of CGI / special effects would go down as tech improved, but it seems like they've gone up instead.

That feeling when you mess up your branding so badly that your "advanced" model gets confused about its own name. #Google #Gemini

Are there any published examples of Double ML or other doubly robust estimators for applied to a ranking problem?

e.g. Predicting the causal effect of displaying a product/post/ad in a particular position of a content feed

I've really been enjoying testing out Pixi for managing software development environments (mostly python, but it also can support non-python dependencies including R packages).It combines a lot of what I like best about brew/conda/poetry.

In case anyone finds it useful, this is a python helper function I wrote that let's you write out the data generating process and get back a dataframe and a DAG representing the causal pathways.