dweez

I spent a lot of time researching LLM routing last year and also came to the conclusion that it's generally not worth the effort. It's too hard to understand the difficulty of a query a priori.

One specific challenge I was seeing is that difficulty depends a lot on what information is retrievable by the agent. Consider the question "what is the 5-state busy beaver number?" (https://en.wikipedia.org/wiki/Busy_beaver). In 2023 this would be a Mythos-tier research problem, but a solution was proved in 2024 so today any minimally intelligent model with a web search tool can just fetch the answer. You don't know which queries will be basic summarization and which will be deep reasoning until you get going.

Good moment to revisit this fun article: https://jasonfantl.com/posts/Universal-Unique-IDs/

If the entire universe were turned into a giant computer and did nothing but generate uuids until its heat death, how many bits would you need for the ID space?

If you follow Apple's official address to a lawyer's office in Delaware, don't be surprised that Tim Cook isn't there to greet you.

The modern world is like Jean Baudrillard's vision of hell. Back in 1991 he wrote "The Gulf War did not take place", commenting on what was at the time a new development of 24/7 live media coverage of the war. Media saturation created a hyperreality where images about the war replaced the thing itself. How far we have come. We are so complacent here that war exists only as stream of symbols and sounds streaming out of our screen. I think many do not truly believe it is real.

If it's any consolation , VTI is free-float weighted so won't pick up very much SpaceX initially.

This is a slightly tongue-in-cheek way of saying that if you believe a security is severely mispriced then there is a straightforward way to express that opinion.

So sounds like this will be a great short candidate after the index re-weighting.

One way to think about this is you have a binomial distribution with p=0.8 and n=number of lying friends. Each time you increase n, you shift the probability mass of the distribution "to the right" but if n is even some of that mass has to land on the "tie" condition.

I wrote a quick colab to help visualize this, adds a little intuition for what's happening: https://colab.research.google.com/drive/1EytLeBfAoOAanVNFnWQ...

There are a few markets like Jacksonville and Atlanta where there is a lot of institutional ownership, but outside of those few cases impact of Blackrock et al on housing markets is effectively nil.

The Agricultural Revolution and its consequences have been a disaster for the human race.

To summarize the article: buying the Yes side of this market is like shorting treasuries. It's not a bet that treasuries will default, but rather a macro bet about that demand for cash (i.e. interest rates) will increase.

Great point. Never forget about counterparty risk!

Matt Levine has a bit about the best customer service your broker can provide is not picking up the phone in a crisis.

Bad UX is, intentionally or not, consistent with Vanguard's long-term index investing philosophy. Call us? Use our website? Whatever it is you are trying to do, you probably shouldn't be doing that.

I kid, but only a little.

Can't get insurance -> can't get a mortgage -> can't buy a house

That had been GitHub's model for their SF HQ office, more or less. Some local people used it as their daily office, but there seemed to always be a mini-summit or get together of remote folks happening each week.

Matt Levine today:

> Surely the highest-variance aspect of the Twitter vs. Musk saga is Zatko’s whistle-blower complaint. If Zatko can make a compelling case that Twitter is horribly bad — that its information security is so bad that it violates the law, that it has fraudulently concealed its problems, etc. — then that is probably Musk’s best argument to get out of the deal: Twitter is doing fraud, it has suffered a material adverse effect, etc. If Zatko is just a run-of-the-mill paranoid security researcher who is aggrieved about being fired and making mountains out of molehills, then his complaint will quickly be kicked out of court and won’t affect the Musk deal. Zatko’s credibility — whether he’s telling the truth, and also whether he is exaggerating or underselling the importance of Twitter’s problems — is a key input into your evaluation of Twitter’s stock value. The more credible he is, the less likely it is that Twitter will get $54.20 per share, and the less Twitter will be worth without Musk’s deal.

> So if you are a hedge fund, or an expert-network firm working on behalf of hedge funds, you obviously want to know how credible he is. You might, for instance, want to talk to some of his old coworkers to get a feel for him. You might offer to pay them a lot of money for a one-hour phone call, because you might have a lot of money riding on the Twitter deal, which means specifically that you have a lot of money riding on your evaluation of Zatko’s credibility.

Oh yeah totally! I want to round up some friends now to play AI Dixit. An easy version could be to play sort of "reverse Dixit" where one person generates an image from a prompt and everyone else comes up with prompts based on the image, then you guess which prompt was the real one.

Okay since we're not going to improve the speed of light any time soon, here's my idea for speeding up CA to NL roundtrip: let's straight shot a cable through the center of the earth.

"The industrial revolution and its consequences have been a disaster for the human race" but make it instagramable