Alex Ames

CFL limit is a real bastard, isn’t it? I spent about 16000 cpu-hours simulating half a nanosecond of an ICF problem in 1D and still didn’t get to convergence

I think something like a right of refusal/bidding process would be a fair way to allow companies to retain IP if they’re willing to pay the higher licensing fee associated with a higher valuation

What? I think you’re reading things that aren’t present in my words.

If Company A is a patent troll and Company B is productive, this is a good outcome. If Company A values the patent at what it’s worth, now they have its worth in dollars instead of in an illiquid patent they’re paying fees on.

If this transfer causes you more than $x harm, then the patent was actually worth more than $x, right?

What’s the mechanism of harm here? You have a patent that’s worth $x, and a company pays you $x for it. You now have money instead of a patent. The company now owns the patent and pays the licensing fee instead of you.

A Harberger Tax on Patents Self-assessing innovation taxes for fun and profit

Intellectual property ownership does not always foster innovation (see: patent trolls). If the ongoing cost of the government-provided monopoly was tied to the value of the idea, patent trolls would go extinct. Harberger taxation! arpitrage.substack.com/p/a-harberge...

Unemployment may increase, but I don't see interest rates dropping significantly given the current levels of publicly-held debt and the great gray wave coming for our old-age safety nets. Workers vote against higher taxes and the elderly vote against reduced benefits, so we'll inflate the debt away

"Make it easier to build things" shouldn't necessitate private funding or ownership. It should mean a given public dollar can build more than it otherwise would. "Who should own things" is separate from the question of "can we build things".

The GND was designed around solving last decade's problems (high unemployment, slow growth) using last decade's tools (cheap debt). The only continuity with present conditions is the continued threat of climate change, but the approach must not lunge on zombie-like from a decade prior

No objections here to building lots of wind and solar, that’ll be the bulk of a clean grid in many places! You just need some (10-40%, depending on regional weather) clean firm generation to back it up. Tech maturation & supply chain development takes time, so let’s start small-scale deployment now

Rayleigh scattering goes as λ⁻⁴, so even if you're blocking out direct sun, UV scattering from the clear blue sky can still do you in eventually

Doesn’t have to be nuclear, necessarily, but wind and solar will not be enough in all geographies. We should make sure we have the tech portfolio needed to completely decarbonize, not just pick the cheap low-hanging fruit and call it good enough
bsky.app/profile/alex...

The shots didn’t actually happen until halfway through my postdoc, but yeah, pretty wild to skip Omega entirely and go straight to NIF! Really not sure what the facility shot allocators were thinking

But we can pat ourselves on the back secure in the knowledge that the theoreticians would be hopelessly lost without experimentalists there to point at expensive data telling them they’re wrong

I spent a few months in grad school designing a small series of NIF shots. My pay for that period was probably less than a hundredth the cost of the facility time for the shots. My own lab at UW used a $300k camera to capture light from a $500k laser. Could’ve funded a small army of theoreticians!

If a system without markets has never been successfully created despite more than a century of devoted effort across dozens of countries, then I’m not going to spend a lot of time yearning for or philosophizing about a spherical cow. You’re welcome to! Doesn’t impact me, everyone has their hobbies

Every existing nominally communist political system has markets, though? The existence of markets is a separate fact from whether the means of production are publicly or privately owned. Every functioning economy has a mixture of public and private ownership with regulation to rein in excesses

Also helpful: large flat commercial/industrial roofs, installation economies of scale, no need to care about aesthetics, lower marketing cost

Also good because the sun’s up during business hours (supply and demand are coincident). The problem with resi solar is that the sun’s setting as people come home from work and turn the AC on, start cooking, plug EVs in, etc., so you need batteries to store solar for later in the day

I loved living in a walkable city as a 20-something, but I never saw kids running around unsupervised. I moved to a SFH-dominated medium-density town for work a couple of years ago, and now I see kids playing unsupervised all the time. It’s much easier to be a parent if your kids can be latchkey

You don’t even need a three-deck edgelord helicarrier for this! Currently-existing carriers already have four catapults

For many technical problems, absolutely, but system cost at high levels of battery-backed VRE penetration is still a stinker! LDES costs are not low enough to handwave away

And decentralization makes it more challenging to ensure everything is actually configured properly. If we’re substituting inverter software for what used to be an inherent property of the grid, that software needs to be vetted well enough that it can be treated like a law of physics

I’m just saying that software isn’t enough alone (or software + solar), you really need physical infrastructure (which could include batteries or synchronous condensers) capable of dumping alternating current onto the grid

If you want to fundamentally rethink how the grid is protected to avoid the need to supply fault current, great, but first we need to figure out how to do that. If it’s possible, then we’ll have to rebuild much of the current grid in addition to all the grid expansion needed to decarbonize

For circuit protection to function properly, you need to be able to supply fault current. You can easily do that with spinning rotors, and more speculatively with big batteries and properly-configured grid-forming inverters, but software’s not enough, you really do need the ability to dump electrons

The branding is still pretty good

blockenheimer.click allows you to block everyone who liked a given post, which can be really powerful if used judiciously

The job of an LLM is to predict the next token. If you ask a human to predict the next word in a sentence, they’ll use their knowledge of language and the outside world to guess. The question is to what degree LLMs develop consistent internal world-models and emergent behavior during training.