The biography of the author is interesting and relevant, but hardly all that matters about this or any other writing.

Yes… but more often something like laser-sintered metal printing, which is not going to melt when hot.

Yep, and it works even in the SpaceX Raptor engine.

We unfortunate know very little about how dementia / Alzheimer's develop in the first place.

Very bad racially-motivated atrocities are perpetrated today. Characterizing that as "slavery never ended in the US" is trivializing a much, much worse past regime.

It is both technically (because of the penal exception in the 13th Amendment) and substantively (mass incarceration and penal slavery were adopted as policy directly and almost immediately as a replacement for chattel slavery, and have spread from their original geographic domain since) accurate, it doesn’t trivialize anything.

The 13th amendment explicitly provides a carve out (“as punishment”) for slavery/involuntary servitude. We still allow slavery in the USA.

This incident took place at a prison, the “as punishment” space that allows the state to enslave American citizens.

One of the great opportunities to improve politics in this country would be to increase our moral dynamic range, to be able to hold in our heads that slavery was very bad, for profit prisons are very bad, and slavery was much worse.

I recommend this Ben Thompson piece on why resiliency has declined: https://stratechery.com/2025/resiliency-and-scale/

I saw these in the basement of a data center about 11-12 years ago. Most steampunk thing I've seen in real life.

Here's a bigger one at ASDEX (a fusion experiment):

https://www.ipp.mpg.de/4244138/generatoren

Likely less than the sum of costs and losses associated with verifying photos of receipts.

My expense card has an app that uses AI to OCR your receipts, automatically fills out 95-100% of the expense form, and for most transactions auto-approves it via per diem policy rules.

An oddly gauzy piece. As an ex-Amazonian, I recommend the (complimentary, insider-written) book "Working Backwards" for those interested in a substantive look at how Amazon ticks.

> At Amazon, customer obsession isn’t just a value—it’s a constraint on every technical tradeoff.

Gauzy because the author simply fed his notes into GPT-4o or 5-instant. If the line above ain't rock-solid proof of this, I don't know what is. And I don't think that our, uh, author gave the model enough to work with.

Didn’t feel like it offered any insights honestly. Guy is feeling holier because he finally gets to work at Amazon.

> unfortunately it kinda shows when you know the material better than he does. (source, I've been lecturing on all of it for a few years now)

That source is bearing a lot of weight.

4 years of going through the algebra of back-propagation with maths and physics undergrads, it's not that difficult :). The main challenge is combining it with stats and almost infinite dimensions of freedom which makes implementation extremely painful. hats off to the guys behind pytorch and tf for making it possible without having to rely on minuit or the promises of minuit2

Not quite; g drops materially between 100 km and 1000 km. 8 minutes I believe is quite close, whereas 5t^2 (or even 4.9t^2) will lead to underestimating time.

The radius of the earth is around 6300km. The difference in g at the start (between 6400km and 7400km) is 25%. But gets less as you fall. So it might make somewhere around a 10-15% difference overall? So s=5t^2 is fine unless you need a super accurate figure, in which case you need to do some calculus. I would trust an LLM with calculus even less.