daniel-thompson

https://www.natesilver.net/

i was a casual reader of 538 back in the day. his substack feels pretty similar, if smaller in scope.

Completely bespoke models are typically trained in Python using tools like TensorFlow, JAX or PyTorch that don't have real non-Python alternatives

The article outlines some interesting ways to evade this problem. What's the latest thinking on robustly addressing it, e.g. are there any approaches for executing inference on a tf or pytorch model from within a golang process, no sidecar required?

The actual paper is hidden behind a Nature paywall. There is a preprint on arxiv here https://arxiv.org/abs/2403.18510

I mean, it _could_ be the job. But a lot of older doctors feel like the younger generation needs to go through the same struggles and hardships they did. This reminds me of something I occasionally see in my parenting of my kids - the desire I have for my kids to do things that validate _my own_ childhood experience.

yet

I think it's a pretty simple, and timeless, aspect of human nature:

"People are generally better persuaded by the reasons which they have themselves discovered than by those which have come into the mind of others."

- Blaise Pascal, 1670

carriers can be used to project power in scenarios short of all-out war, which is the common case, at the moment

and cheating

one of my favorite pieces of financial journalism ever, https://www.bloomberg.com/opinion/newsletters/2019-10-23/mon..., explains how he got the bux during wework's collapse:

[softbank's offer to rescue wework] is the result of Neumann’s holdup power: Prior to the deal, Neumann is still the company’s controlling shareholder, and he could just say no to a deal that he didn’t like. That might completely evaporate his own wealth, but it would evaporate a whole lot more of SoftBank’s, and it kind of looks like SoftBank blinked first: In effect, the price of Neumann allowing SoftBank to rescue WeWork was that SoftBank had to hand Neumann a billion dollars for himself

i've been reading this blog off and on for years, it's fantastic. a good introductory entry for those not familiar with the concept of a mcmansion: https://mcmansionhell.com/post/149284377161/mansionvsmcmansi...

https://en.wikipedia.org/wiki/Stellarium_(software)#Screensh...

like-new restoration of various old objects: - https://www.youtube.com/@mymechanics

annotated reviews of high-level chess games: - https://www.youtube.com/@ChessNetwork

blender tutorials: - https://www.youtube.com/@blenderguru

rules for thee, but not for me

"perfect synchrony" is kind of meaningless, might be better and more accurately titled as "first solar system with a six-planet orbital resonance has been found"

I didn't say it was a surprise; I said the marketing claim made me laugh.

I know Docker isn't something Apple makes, but still, this makes me laugh. I run a fair number of containers as part of my daily workflow and the memory overhead, compared to a generic linux amd64 box running the same thing, is on the order of 10x. So for me it's actually more like, 8GB on an M machine == 1GB "regular" RAM.

Can you elaborate on this a bit? I'm not too familiar with this scene but I'd like to get one, and not the crappy $14.99 version at the local big box retailer.

13K commits over the past year works out to almost 40 commits per day. Curious what the workflow is to accomplish that and the nature of the commits. Irrespective of that, that is super intense, sounds like she's getting a well-deserved rest.

This is just terrifying. The only thing I can imagine worse than having it myself would be my child having it. Nightmare fuel.

From https://github.com/ammario/redjet#streaming

  To minimize allocations, call (*Result).WriteTo instead of (*Result).Bytes. WriteTo streams the response directly to an io.Writer such as a file or HTTP response.

Wrong title - this is James Cameron's speculation based on unnamed sources.

This is really cool! But unfortunately for me, it gives me vertigo after a minute or two. Anyone else have that sensation and/or any pointers to resolve? I'd like to keep playing with this if possible.

The author demonstrates attaching services to a context, which afaict goes against the Go authors' usage recommendations in https://pkg.go.dev/context:

Use context Values only for request-scoped data that transits processes and APIs, not for passing optional parameters to functions.

I'm no astrophysicist, but the paper and some web data show

- Sag A* has a mass of ~4,000,000 times the sun, and an event horizon of approximately 12,000,000 km (0.08 AU). This is just the event horizon; Sag A*'s accretion disk has a diameter of about 150,000,000,000 km (1000 AU). So your intuition would certainly be right about Sag A* - there is a huge disk of gas and other junk you'd be flying through.

- This object has a mass of at most ~12 times the sun, and (if it's a black hole) an event horizon of approximately 35 km. The paper doesn't say anything about an accretion disk given that it's exploring the idea that the object is not a black hole. Regardless, a star orbiting at 1.4 AU would probably clear out everything in the immediate neighborhood.

More likely != confirmed

[1] S62 comes within 16 AU at closest approach to Sag A* but its orbit is highly elliptical.

The distance between objects in this system 1.4 AU, which is _incredibly_ close.

https://en.wikipedia.org/wiki/S62_(star)

How does [person A choosing not to speak at a conference] connect to [person B leaving the Rust project]?

Dawson's law strikes again!

O(n^2) is the sweet spot of badly scaling algorithms: fast enough to make it into production, but slow enough to make things fall down once it gets there.

https://randomascii.wordpress.com/2021/02/16/arranging-invis...

When I was in high school I could ace most of it without studying at all. I thought it was great at the time, but the shitty part - only apparent later - was that I failed to develop good study habits.

When I started at university in a technical major, it was a real kick in the teeth. The material was substantially more difficult, but the hardest part for me - and the biggest piece of learning - was the realization that I had to study consistently and be purposeful about managing my time in a way I never really had to before I got there.

My grades freshman year were abysmal. But they also helped me internalize that I needed to buckle down and put in the work. I'm glad that that happened in a relatively forgiving environment like school, instead of in the workplace.

Local LLMs do best on big GPUs, which aren’t an option on Macs.

For processing speed, Ms are fast but I agree not anywhere near a top Nvdia chip.

For memory size, the memory on an M chip can be used as graphics memory, so a person could get an M2 today with 128GB of graphics memory for ~ $5k. Not bad considering an Nvidia chip approaching that memory size is several times that much.