Keunwoo Lee

@Ussjoin@cloudisland.nz @lcamtuf@infosec.exchange @simplenomad@rigor-mortis.nmrc.org I believe this piece of writing was substantially LLM-assisted

@lcamtuf@infosec.exchange a number of years ago Stefan Savage of UCSD was researching the spam ecosystem and somehow got approval to spend grant money ordering Viagra over the Internet. He got the pills. https://cseweb.ucsd.edu/~savage/papers/Oakland11.pdf

A couple of prominent anti-AI intellectuals recently did a podcast aiming to show that AI's alleged utility in software engineering is "bullshit". I went in hoping to hear the strongest arguments for this view (which would contradict my lived experience). The podcast mostly consisted of the hosts pointing at Garry Tan's github repo and smirking. I think Tan is an asshole and an idiot but this episode reinforced my impression that the anti-AI discourse is mostly extremely low quality.

Apparently somebody is floating the idea of making Venezuela a US state. I like this idea so much that I propose making 150 new US states out of Latin America, Canada, and let's just throw Greenland in too. The new states can hold elections immediately for their Congressional seats, then ratify amendments that rewrite the Constitution and mandate a special Presidential election. It's not truly the United States of America until we get all the Americas in. 200 STATES OR BUST.

Unfortunately, given the way the AI culture war is shaking out, I expect that a lot of companies that embrace agentic work are going to undervalue, alienate, and drive away their Togusas, or perhaps just fire them outright.

Lately I've been thinking about this character a lot.
https://en.wikipedia.org/wiki/Togusa
When I first saw Ghost in the Shell in the 1990s, I thought the premise was silly and that, in practice, the guy with the antique revolver and no cybernetic enhancements would just slow everyone down and eventually get slaughtered. Now I think the Major was right: every engineering team practicing agentic development needs a Togusa, for cognitive diversity and as a backstop against cybernetic failure modes.

@danluu Aren't we just talking about why humans find it hard to delay gratification? Of course good testing, like many kinds of up front investment, can save time in the long run; the "long run" here might even be next week or sooner! But not writing the test lets you push to main right now, so you grab the marshmallow.

American observers say that if we don't push the AI frontier forward, then China will. I take a dim view of the Chinese government's attitude towards individual liberty and many other issues (note that I could say the same about the current US government), but Americans should grapple with the idea that maybe China taking the lead in the AI race wouldn't be so bad.

China is also much less a petrostate than the US and its leadership class has no vested interest in delaying the green energy transition. On the contrary, its leaders know that due to geography, the Chinese population is more exposed to risk from global heating than the US, and China has been the the engine of the past decade of global energy progress. Chinese AI will eventually be entirely powered by solar and batteries.

A curious blind spot for AI doom forecasters in the US is forgetting that China is, at the top, a command economy. US AI may blindly hyperoptimize for the needs of capital but the CCP has no ideological commitment to the independence of capital per se, and unlike in the US you can easily imagine the state de facto nationalizing AI and distributing a lot of the value it generates via social benefit programs. This also applies to some degree in Europe, though Europe will lag in development.

Long form text with sections, paragraphs, and varied sentence structure was briefly a signal of quality for online writing (shallow and fallible, but still nonzero signal). LLMs have completely blown this away. What you want now is a terse brain dump by someone who has a visceral distaste for wasting your time. Unfortunately such writing will necessarily be more esoteric, but I guess those are the breaks.

the other categories are reasonably self-explanatory, but I should clarify that by "thugs" I mean authoritarians and their followers. thuggery and bigotry are natural companions: the bigot provides the target against whom the thug's violence can be deployed, and the thug provides the means by which the bigot's subjugation of the other can be operationalized. but the thuggish impulse precedes the target, and is distinct from it.

i'm not just hurling insults around; when I say this model is functional, I mean that you can make sense of Republican behavior as emergent from the interplay between these tendencies. the bigots and thugs have always been there, but the reason they're ascendant now is that long-term economic trends have made it necessary for plutocrats to cynically whip up anti-immigrant/anti-LGBTQ sentiment to distract people.

the most functional model of the US Republican Party (not just now, but my entire life, & I'm in my 40s) is that there are 5 main tendencies: plutocrats, theocrats, bigots, thugs, and marks. any given Republican is some blend of the 5. the party's behavior varies depending on which faction is ascendant. when I was younger plutocrats & theocrats mostly ran the show; right now it's thugs and bigots, with a side helping of plutocrats.