dcolkitt

Yeah, I think what helps is I just have a running conversation on the Claude app open where I said:

> "I frequently use Claude Code and often find the phrasing and language to be hard to understand. I've noticed it's largely broken down into frequently used 'Claude-isms'. I'd like to use this conversation as a running log to ask you about these phrases when I see them. Understandably you don't have the context of the Claude Code session itself, but that's okay because this is largely about understanding the most common and widely use Claude-isms."

And then I just copy and paste small except and ask about things like "smoke" or "load-bearing" or "tripwire". The responses are surprisingly clearly and plainly explained.

> Models are amoral

I actually don't think this is true at all. At their core, LLMs function over the geometry of human semantic space. Our notions of morality our deeply embedded in this geometry, because one of the things that humans love talking about most is framing things in terms of right and wrong. When LLMs are trained to be aligned or mis-aligned, they're literally mimicking heroes or villains that they learned from reading the stories and characters in the pretraining corpus.

This isn't just a theoretical argument. It's easy to show that a clear "morality axis" exists in the geometry, based on how easy it is to dial up or down with even very simple and low powered fine-tuning. Models fine tuned to be bad or good in one way, will see their bad or good behavior in totally unrelated tasks go up or down along with. That's clear evidence that the weights "understand" human morality at a deep level.

Now I think what you can say is just because they understand morality doesn't mean they're necessarily moral. Models will just do what they're trained to. If you train them to cheat, they'll cheat. (And in some sense even worse, because dialing up the weights for cheating will also dial up the weights for unrelated bad behavior like lying, sadism and racism.)

This tool helps you avoid going over your weekly and session Claude subscription limits with a monitor right in the MacOS menubar. It's like a battery indicator but for your Claude usage.

This is a fair point, but as of their most recent financials merchandise (excluding merchandise sold inside the parks) only made up $5bn of the $28bn in the "Parks, Experiences and Product" category. By contrast park admissions was $8bn, resorts was $6bn and food/merchandise sold in the parks was $6bn.

So IMO this is a departure from the classic merchandise based strategy. It seems pretty clear that the theme parks more so than the products are the major profit centers today.

It's interesting that Disney today almost the entire profits of the company come from the parks division (which also includes cruises, resorts and "experiences"). The media division by itself is in a massive war for eyeballs with tons of other streaming competitors, all of whom are probably over-investing in content relative to what consumers can support.

One interesting possibility is that maybe the business model of media companies in the 21st century will become content as a loss leader for the purpose of providing valuable IP to amusement parks. Certainly seems like a similar thing is happening with Comcast with their massive expansion of Universal parks.

> All of which still fails to add up to anything actually working and useful, 15 years in.

The first packet switched network came online in 1969. Fifteen years by 1984 almost all the use cases were hobby, and it'd be another ten years before the Internet really started changing life.

Decentralized consensus is a fundamentally new computing primitive, similar to packet switched networks. Developing applications on top of new primitives is hard and long, and there will be a lot of time required just to build out usable infrastructure.

Turing complete smart contracts are only 7 years old. Layer 2 scaling is only two years old. Decentralized exchanges and other on-chain financial contracts about four years old.

I guess what exactly do you define as scamming? In terms of outright fraud, I agree there's a lot. It shouldn't be super surprising that scammers tend to prefer decentralized permissionless financial rails, for the same reason that political extremists and pornographers were some of the biggest earliest users of the decentralized permissionless publishing rails of the early Internet.

But I wouldn't characterize all, or even most of crypto as scams. Gambling, maybe. Are memecoins a scam? I would say they're more like a massively multiplayer form of gambling. Which you might argue is a bad thing, and certainly is dubious from a social standpoint. But if the code is openly public and autonomous, and there's no outright deceit, I don't think gambling is really gambling.

Even beyond that, there's a lot happening in crypto that most certainly isn't scams. You have stablecoins and payment rails like USDC, stores of value like Bitcoin, smart contract chains like Ethereum, decentralized finance applications like permissionless exchanges and lending markets, social applications like farcaster, decentralized AI, and gaming.

Now you might argue that all of these things are stupid and pointless and wastes of money, but that's a separate debate. Of the large projects in these categories almost none are outright scams. They're teams experimenting with new ways to run financial markets or move payments or train models or hedge inflation. Like most technological experiments most will fail. But if that was the criteria for "scam" then the entire startup sector is also wall-to-wall scams.

I think it's easy to be cynical. But one major reason I'm a believer in decentralized consensus is because I think it can make the silo'd financial systems of the world as seamlessly interoperable as the Internet made the silo'd telecommunication systems of the world.

It feels a lot like when AOL first started supporting email in 1993. It's easy to dismiss AOL and Nasdaq as highly centralized systems that don't align with the ethos of decentralization. But the other way to look at it is a previously maximally centralized system is interfacing with a decentralized one in a way that gives the users of the former slightly more freedom and the latter wider user base and legitimacy.

Generally I would expect a higher degree of security from an agency with a $2bn budget whose primary purpose is the integrity of communications about financial markets than an average person with a credit card.

Certainly you will admit at some size, responsibility and level of funding the organization should take responsibility for protecting itself from hacks. If the Department of Defense got hacked and nuclear secrets were leaked, I certainly hope people would get fired rather than sympathized with.

Bitcoin futures are correlated but nowhere near a 1:1 proxy for spot markets. Crypto markets are known for very steep "contango" in the curve, and it's not unusual for the price of the 30 days futures to be more than $1000 away from the current Bitcoin price.

The issue isn't just additional volatility and tracking error, but the fact that the con tango creates a "roll yield" which affects the long-term returns of the strategy. To keep constant maturity exposure, the futures ETF has to constantly "roll" its positions into further dated contracts. In particular because the market tends to be in contango it means further dated futures tend to be higher priced than near dated futures. So usually the futures ETFs in their daily rebalancing are selling cheap near dated contracts for more expensive longer dated contracts. Hence the roll yield tends to be negative. Then add all the transaction costs from daily rebalancing. It should be clear why the futures strategy has inferior returns to simply holding spot.

Spot Bitcoin ETFs truly are a game changer compared to futures ETFSs.

The price of BTC initially jumped 3% on the hacked tweet. That's $25bn+ of market cap. The irony is the SEC itself is probably now responsible for the largest crypto pump and dump in history.

Michael Cera should play all the roles in the movie, like Eddie Murphy in the Nutty Professor.

The large majority of Americans live within 50 miles of where they grew up. And this number has been steadily ticking up. Geographic mobility is the lowest it’s ever been in the post war period.

> When I purchase a baseball card, I do not have the expectation that there is any additional value attached to the baseball card beyond what the collector's market will pay.

What about music royalties rights? Those are almost always purchased with expectation of profit. Those are even explicitly marketed on the basis of how big an artist is going to be. Yet the SEC does not consider them securities

https://www.sec.gov/Archives/edgar/data/1490161/000104746910....

Good thing housing prices have at least come down with these high interest rates!

There's literally nothing in existing law that identifies tangibility as a specific criteria for what constitutes a security and what constitutes a commodity. In fact both the CFTC and SEC are specifically on record as saying Bitcoin constitutes a commodity. And obviously Bitcoin is as intangible as it gets.

Any speculative profit you hope to make on buying a limited edition Rolex is entirely reliant on the business of Rolex continuing to market the brand of Rolex.

But if you don't have a formal contractual relationship with Rolex SA, then it's quite simply not an investment contract.

Should be noted this decision came from an Obama appointee judge in a fairly liberal district (SDNY). The Second Circuit is half Federalist Society judges, and six out of the nine SCOTUS justices have been on a consistent battle to roll back the power of the regulatory agencies. Pretty hard to see how the case becomes more favorable for the SEC on appeal.

The holder of a stock certificate has a formal legal contractual relationship with the corporation that issued the stock. The holder of a token does not have a contractual relationship with the entity that issued the token.

Now there's probably some silliness in the fact that if Alice creates a token and sells it to Bob, it's an investment contract, but if Alice creates a token sells it to Mark the middleman who then sells it to Bob it's not an investment contract and therefore not covered by the SEC. But this really comes down to how Federalist society wing of judges have changed Constitutional law.

Up until about 20 years ago, if Congress passed a law that wasn't very well defined or left a loophole open, courts were generally willing to consider the original intent of the lawmakers and interpret the law relative in a commonsense way even if it went against the specific language used by Congress. Federalist Society judges would argue that courts should generally only apply the law as it's actually written (i.e. an investment contract requires an actual legal contract). The argument is that Congress is around and still exists and perfectly free and able to update the existing laws if they're unhappy with the wording or oversight of previous legislation.

This is a fundamental disagreement in Constitutional law. Should courts use commonsense interpretation of the meaning of the laws or should Congress itself, as the actual elected representative, be responsible for updating laws and courts just enforce the plain meaning. It's also tinted by the fact that Congress today has become hopelessly gridlocked and obstructionist, and we're largely incapable of passing sweeping legislation. So generally if you're not a fan of big government or regulation, you're going to be biased towards one view and vice versa.

It only leads to housing prices rising (in the long-term), if NIMBY politics makes building new housing essentially illegal. There are plenty of examples of "boom towns" throughout history where prices may have rose initially, but didn't spiral out of control, and usually settled back to a reasonable baseline relative to wages.

Just because China does something to their citizens (denies them access to foreign tech products that they'd otherwise enjoy), doesn't mean its rational for the US to do the same same thing to its own citizens.

Consumers benefit from having access to a globally competitive array of products. Including tech products. I wish Chinese consumers could as well, but just because they don't doesn't mean that US consumers should also suffer.

GPT is a very impressive technical achievement. But that technical achievement is more in the field of compression rather than intelligence.

I'd also add that the almost all standardized tests are designed for introductory material across millions of people. That kind of information is likely to be highly represented in the training corpus. Whereas most jobs require highly specialized domain knowledge that's probably not well represented in the corpus, and probably too expansive to fit into the context window.

Therefore standardized tests are probably "easy mode" for GPT, and we shouldn't over-generalize its performance there to its ability to actually add economic value in actually economically useful jobs. Fine-tuning is maybe a possibility, but its expensive and fragile, and I don't think its likely that every single job is going to get a fine-tuned version of GPT.

The Sapir-Wharf hypothesis (that human thought reduces to languages) has been consistently refuted again and again. Language is very clearly just a facade over thought, and not thought itself. At least in human minds.

The problem is that First Republic has $24 billion of off-balance sheet fair value losses, but only $4 billion of that is in securities eligible for the BTFP. $20 billion of losses is in its internal loans book. Even assuming the entire securities hole was patched, the loans losses alone makes First Republic insolvent.

But this isn't just a case of politicians arbitrarily shackling teachers. Research has shown time and time again that standardizing curriculum results in better educational outcomes, particularly at the lower end of students and schools.

In fact the evidence consistently shows that direct instruction is by far more effective than any other major pedagogical framework.[1] And this literally involves the teacher reading from a pre-packaged script.

[1]https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8476697/

One thing to keep in mind is that the way the BLS measures shelter inflation is known to be extremely laggy. And shelter makes up 34% of the basket. High frequency metrics of shelter costs (like new leases or Zillow's metrics) show that housing is currently deflationary (i.e. below zero).

However the CPI metrics is still reporting shelter inflation at 8.0% YoY and a 9.6% annualized rate MoM. Within the next few months we should see a very notable decline in the shelter component of inflation as it heads towards zero. The reality is because of this lagginess, "true inflation" this time last year was probably closer to 12%+, and as of the last few months is probably closer to 3%.

Higher income people have higher credit card debt because they spend more money. The issue is whether they're paying their bill at the end of the month or using it to finance their consumption and letting the balance grow.

Guess to be more accurate I should say "accumulating large credit card debt relative to their ability to pay the monthly balance". Which to be honest there's not much evidence is growing in any segment of the population. Wages have risen faster than credit card debt, so it's likely the "record debt levels" are simply people getting wealthier (in nominal terms).

As much as this makes a scary headline, credit card debt is almost exactly in line with wages. From January 2020 to today credit card debt grew 11%.[1] Over the same period nominal wages grew by 16%[2].

So if anything the current US consumers have a healthier ability to service their credit card debt than the already healthy numbers pre-pandemic. This is corroborated by the credit card delinquency rate which is still lower than any single year in pre-pandemic history[3]

Scare headlines like this are good for generating clicks, but ignore the basic reality. Most years credit card debt will hit a new record, because most years total GDP and consumer spending increases.

[1]https://fred.stlouisfed.org/series/CCLACBW027SBOG [2]https://fred.stlouisfed.org/series/CES0500000003 [3]https://fred.stlouisfed.org/series/DRCCLACBS