"The Apple of Loving Grace" -GPT (made with AI)

Imagine thinking that Geoff Hinton invented LLMs...

This analysis is spot on. The real H1b problem is almost entirely in her category 3 (non-tech companies huring te h workers), which is also likely the vast majority of H1bs in the entire country. I work mainly in startups, and the H1bs I interact with there are outstanding. But most H1bs are working for companies like Walmart or Disney, and they are doing jobs that could easily be done by Americans from any part of the country, jobs that are frankly not that difficult and only require an IQ of 120 or so.

Individuals differ greatly from population averages, but stereotypes have many avatars.

Does anyone else think need a better notation for linear tensor operations? Reading papers without the code it's sometimes hard to see what the operation actually was. Does this bother anyone else? I suggest we should do something based on einsum notation, e.g. $\left[\underset{a}{ijk}\underset{b}{kn}\right]_{in}$ for $\sum_{j,k} a_{ijk} b_{kn}$.

Yep, though it's more like a century and a half (to the 1870's) than just over a century.

So the essence of this story is that if you pay people on the internet to take surveys, somebody somewhere on the planet will figure out how to grift some money for fake results. That, and that university administrators at Harvard are horribly credulous investigators. That, and that it really doesn't take particularly good ideas to get tenure at Harvard. "People who act inauthentically feel the need to purify themselves afterwards" ... you call that _research_?

The cold start problem is one of the main issues with RL historically. You can reinforce action trajectories until you see them. The other big problem is credit assignment; not all actions share equally in a reward, and RL rewards all of them anyways.

Next time someone tells you something will "definitely" happen, keep in mind they only mean it's got about a 78% chance. Conversely, if someone tells you it's highly unlikely they'll do something, there's still a 17% chance they'll do it. People are weird.

If this was your grandmother, I'm very sorry for your loss. For every else ... here is the modernized version of "Grandma got run over by a reindeer" ...

What a weird take. In reality, wealth is goods and services. In a healthy, functioning market system, money primarily enables the exchange of goods and services. You make money by creating goods or services. So if someone makes a lot of money in a capitalist system, it should mean that person created a lot of goods or services. So this whole idea of a fixed pool of value is just kind of silly. You can make more value by providing more services. Nothing stops other people from ALSO generating goods and services and making money EXCEPT for access to base resources, and that isn't as big of a factor as people like to make it out to be in most cases. Mostly, value is created in services more so than in goods, so wealth is really about the value of labor. Marx argued that capital would conspire to replace labor with machines, but that just misplaces the argument, because in a society like ours you can find people to help you purchase most machines if you're not a total social misfit. The right conversation is "who's blocking your ability to create goods and services", but that has only a tangential connection to "capitalism", and it's most certainly not fixed by "socialism", which puts even more constraints on people's ability to create wealth.

Nice work that trades training for search in latent space in order to generate images.

The brain-as-computer analogy definitely has not been proven, but it hasn't been disproven either. The argument that just because we haven't done something we can't do in the future it is a fallacy. And the argument that we can't create something we don't understand is not a strong claim either.

I never knew that arXiv was moderated until just now...

I recently read through the Three-Body Problem trilogy by Cixin Liu, and I can understand why Obama liked it so much. The tone is pessimistic. The humans are hopeless losers who have trouble doing anything pessimistic, who ultimately suffer an inevitable doom. The humans are so devoted to compassion and protecting every human life that they get pretty much everyone killed by shutting off development, and somehow that's supposed to be viewed as a positive human trait. So, in other words, Cixin Liu and Barack Obama share the same negative outlook on humanity and our future. I can't recommend the book, as it's disjointed and self-contradictory throughout. Major plotlines develop around claims about reality that are somehow ignored shortly thereafter. Like, if the first thing any civilization does when it encounters another civilization is to destroy it without establishing contact, then (a) how is the whole first book about ongoing contact between an alien civilization and earth? and (b) how are there "trade routes" and concerns about interspecies etiquette in the third book? And that's just one example. The second book in the series was pretty good, but overall, not worth the time.