trashedcoast

That's the sort of direct engagement with ideas that would, in itself, be productive!

Like... when Bouie criticizes you, he directly cites passages from your book. When you criticize Bouie, you criticize his tone and the conversation around him. These are not equivalent types of criticisms. An inability to distinguish structural differences between arguments defines so many pundits.

By and large, Jamelle Bouie is good because he directly engages with the arguments of critics. Bouie's critics are bad because they think engaging with the meta-commentary around Bouie is the equivalent of directly engaging with the arguments he makes.

🫡

How would the state identify code written by a local LLM as opposed to a person? How would it identify an image created by a local diffusion model vs something created in photoshop?

How is the state supposed to determine whether a certain type of computation has taken place in order to intervene?

I know you don't care what the people who work with the tech think it can do. That's the whole problem! This is not a case where "how the tech works and what it can do" can be handwaved away if you actually want to implement meaningful policy interventions.

I would maybe take some time to think about what makes regulating stabbing feasible, and how those attributes don't map onto regulating what types of computation can occur.

I'd be more on board with the "ban linear algebra" people if they stopped framing it as a policy preference and instead framed it as a direct assault on God's mathematical fortress.

This is the type of take you come up with if you refuse to actually engage with what the technology is.

It's also abundantly clear that whoever made this video has never ridden the Dubai metro. I have never seen able bodied adults shove people out of the way to claim a seat on any other metro system I've ridden.

Goddammit I made it 2.5 hours into my day without finding out we were at war. It was such an optimistic morning.

it's also notable that historians, scientists, non-fiction authors have thrived here while journalists often *haven't*. I think there are many reasons for that that are not journalists' fault, and some that are.

Sometimes something awful from Online breaks containment into IRL, and sometimes it's important to let an IRL nightmare loose on the TL

A gas station pump ad for chewing tobacco in Montana said to me "it lasts longer than your mom's boyfriend".

Virtue signaling is good because it's a way to signal what your virtues are

Why does the architecture in Spokane go so gd hard?

If you could take some sort of software, let it map the most important statistical connections between everything written on a social media platform, then tried to predict words based on those statistical connections, I bet you could decipher some pretty interesting things.

On the rare occasions I descend into facebook, it's almost like its algorithm is mocking me in the way it chooses to show me exclusively the worst things on earth.

It's sadder because I'm almost kind of sympathetic to facebook's core appeal. I would love a centralized place to stay up-to-date with friends and family. But facebook clearly made a conscious decision to stop caring whether it was even a bare minimum pleasant place to spend time.

I would love to read an in-depth account of wtf happened to Facebook. Like, how did a company let its flagship product rot on the vine so thoroughly?

Deriving Neural Scaling Laws from the statistics of natural language Despite the fact that experimental neural scaling laws have substantially guided empirical progress in large-scale machine learning, no existing theory can quantitatively predict the exponents of thes...

The scaling laws for LLMs are truly scale invariant. LLM performance will continue to improve as purely a function of training data quality and compute. If there is a bubble that pops, it's not going to be because the models stopped getting better.

arxiv.org/abs/2602.07488

In an uncertain environment, interest and curiosity are THE marketable skills. It's how you differentiate yourself, and it doesn't depend on the direction the job market takes. And if entry level jobs don't exist in 4 years, then it doesn't really matter what major you pick.

Nobody has the answer for what the entry level job market looks like in ~4 years. It's fundamentally uncertain. The right question isn't, "what major is the most marketable 4 years from now?", it's "what skills are the most important for navigating change and uncertainty?".

Whatever they're interested in. Nobody really knows what AI's impact on the job market will be. But we do know that a person interested in what they're doing has a leg up on someone doing something they're not interested in.

It's not that simple. Sometimes spreading a meme is as important as creating it. There are many factors.

This is um... wild

I mean... it makes sense that that's the case, but on the other hand, it absolutely does not make sense that that's the case.

It's wild how, for basically every benchmark, once models start hitting 5%-10% accuracy you can just extrapolate the sigmoid to get a pretty accurate timeline to saturation.

Yeah, I support Newsom. I appreciate the harp, and I think she and Andy Samberg are just such a power couple