ThrustVectoring

The real strength of vim (and emacs, though I haven't learned emacs to know this firsthand) is the expressivity of it rather than the raw speed. When you can express an edit as a sequence of key presses, you can record and replay that to easily and make the "same edit" in many places.

The funniest time I did this was with a interview screening question which had a hundred-deep chain of links with the URL for the next link embedded in some JSON with some additional formatting required to get a working link. Vim let me copy a line of text that's a curl command, shell out to replace it with the ~6 lines of JSON response, edit that as a string down to a curl command to the next URL, and wrap that entire process up as a macro to run a hundred times.

Netflix France would be able to pay nearly whatever content licensing and technical access fees it wants to Netflix. It's can make approximately zero profit and need not have any French employees, merely be a container for having a total top-line revenue equal to Netflix's revenues from France.

What's to stop Netflix from spinning out a Netflix France for the sole purpose of having all French revenue in a single corporate entity subject to the 2% of total global revenues rule?

Even if it doesn't, the effect of so much writing getting pushed out the door through default LLM settings is going to make it feel same-y and bland in short order.

Human writers also use specific constructions for definite (though not always consciously known) reasons, and mapping those correlations onto writing models is only getting done on accident. This is a big part of why LLMs doing pointless negations (it's not X, it's Y) is so jarring: people use that construction in order to preempt an expected misconception, whereas LLMs try to use linear algebra over the learning corpus to generate it in appropriate quantities and places according to RLHF/fine-tuning/etc.

In other words, it isn't just verbiage that has a "preferred" quantity, but also higher-level language features, and the appropriate contexts for those features feel like the sort of thing that will be extraordinarily difficult to replicate without a theory of mind.

edit after coming back and re-reading: the "in other words," and rephrase is because I'm trying to explain something complex enough that I'm not even sure I fully understand it, so I try typing out multiple explanations and leave the ones that make the most sense to me. This might be idiosyncratic to my writing and not a significant feature of training corpuses, but even if it wasn't it's a second example of the sort of thing I'd suspect LLMs to be bad at using appropriately.

The critique of AI systems that lives rent-free in my head is that they "hit the target but miss the point." In writing, especially technical writing, the target is conciseness, word choice, tone, etc, while the point is to effectively communicate authorial intent.

https://en.wikipedia.org/wiki/Big_Mac_Index is the closest i can think of offhand

The introduction of rabbits to England post-dates the Norman conquest.

If they're losing money at market rents, what are they going to do about it? If they could raise prices, they would've raised prices before the tax because that's just leaving money on the table. If they throw an emotional fit and leave the unit empty, they're losing even more money. If they sell or go bankrupt, the new owner will have bought it at a lower price and so end up having a profitable business at market rents.

New construction is like 1-2% of total housing stock annually. It'll take a relatively long period of time for these sorts of changes to hit prices because again, low price elasticity of supply.

edit: on slightly further thought, you'd also need this effect to either increase overall vacancy or reduce total construction, since any supply shift from occupied rental units to owner-occupied condos also implies a demand shift from renting to owning.

Landlords charge what the market will bear and generally have little if any price elasticity of supply, so they bear nearly 100% of the economic incidence of the tax. This is especially true where housing supply is constrained by zoning policy rather than the price of expanding supply.

The point is that the most efficient ways to physically handle oil products happens when prices are stable and predictable. I could have just as easily talked about the expenses involved in keeping a profitable-only-at-higher-price field on operational standby as a real option on oil prices, or increased shipping costs as localized price changes shuffle trade routes, etc.

It depends on the price elasticity of demand. When it's 1 - that is, when a 1% change in price results in a -1% change in demand - then any change in supply reduces the total amount of money spent on the good. If demand is more inelastic, then increasing price will increase total spending. Oil is a particularly inelastic good.

Note that the amounts of money involved here are not equal, companies respond to price movements and expected price volatility with less efficient behavior, so the volatility itself causes economic losses.

Like, there's a trade you can do where you load up an actual tanker with oil, park it, and sell an option to buy that oil. The cost of using this tanker and holding this oil a pure waste compared to just having a market-clearing quantity available at a consistent price at all times, but if the market is scared enough it makes money.

Roko's Basilisk was never intended as a serious concern. It's a hypothetical used to verify that a decision-making algorithm does what you want it to - if it suggests that you should get blackmailed by it, you messed something up and need to go back to the drawing board.

You can safely ignore anyone concerned with Roko's Basilisk on topics related to decision theory as someone who is deeply confused about things.

Antarctica has zero permanent inhabitants

They're hamstrung by property tax increases being massively politically unpopular. Prop 13 only bans taxes on the value of land, there's nothing stopping politicians from approximating LVT by an "urban acreage tax" based off plot size in cities other than the fact that it'd get them all voted out.

Electing a censor doesn't make them any less of a censor.

Companies like arbitration because unlike in a civil lawsuit, the plaintiff cannot compel discovery. Paying for an arbitrator is cheaper than going through discovery, especially if the discovery unearths facts materially damaging to the business.

Solar panels don't have good enough power to weight ratios and aren't likely to contribute effectively.

As far as sails go, you need a second medium to act as a keel, otherwise it's impossible to get a net push in any direction other than downwind.

AIUI the problem is the economics are pretty unfavorable for smaller-scale prototypes due to square-cube issues (drag and thus fuel costs is proportional to cross-sectional area, lift and thus cargo capacity is proportional to volume). You'd need a very big airship to get a competitive economic niche between sea and traditional air freight, where you're faster and more flexible than sea transport and cheaper than traditional air freight.

The intent to communicate particular ideas is actually important to the writing. If you have a lot of well put together words with relatively low intentionality behind it, what you have is something that is very effective at wasting your time. If I want to waste my time I'd rather be playing videogames.

Unironically, for a small enough business, 10% of revenue is cheaper than figuring out what the rules are and following them. If I'm doing $30 worth of sales to Malta, it's cheaper to pay them $3, basically regardless of what Malta-specific regulations there are.

The line is between processes you can stop by hauling someone into court and coercing them into stopping things, and ones you can't. Think of a classical computer virus that infects machines and uses the compute and communications to infect other machines - no matter who you haul into court, you have to go and remove it from every involved machine in order to make it stop doing things.

This category of "rogue AIs" are essentially just computer viruses that infect machines by paying to rent them and uses their compute and communications to do various economic and/or criminal activities to get more money to pay to rent machines.

That's because they are extraordinarily suitable as covert torture devices. When government officials use them to inflict pain, they do not leave physical evidence and the only video evidence from filming the interaction is how long hands were on the detainee.

Intent to access a computer would have to be proven for that section of the CFAA to be relevant. The shakedown would be covered under subsection 7, governing communicating threats of computer damage or unauthorized access with the intent to extort.

That actually is how the law works. You can read the Computer Fraud and Abuse Act at https://www.law.cornell.edu/uscode/text/18/1030 and double check, but these felonies all require knowingly or intentionally accessing a computer etc. These aren't strict liability statutes - the government must prove mens rea to a jury in order to get a conviction at trial.

The issue is not a localized part of any particular piece of prose, so being hard to articulate is unsurprising. Even the most egregious of LLMisms are little more than known-likely crutches that the statistics spit out in a sweet spot that gets noticed without being so frequent as to get RLHFed out of the model.

When you speed read, you can grab the words off the page faster than you can understand and fully process the information conveyed by the text. How much time you have or want to spend re-reading or thinking about what you've read is quite obvious. There's a stark experiential difference between reading an informationally-dense passage and one that spends a lot of time rephrasing things, using LLMisms to restate concepts, adding in extra connecting phrases, etc.

If your reading speed is limited by how quickly you can subvocalize the words to yourself, this is significantly less obvious. Unless the passage is dense enough to require multiple read-throughs at conversational reading pace or vapid enough to be boring, you're going to feel done with the text at roughly the same time. Speed readers do a lot more re-reading and varying of reading speed, and that is going to correlate pretty hard with information density.

Google pays third-party websites to host advertisements, and shady website owners will put ads on such a website and bot clicks on it.