ucha

> “By definition it must make text worse … because the nature of the watermarking algorithm requires it to sometimes increase the probability of selecting a worse word choice and decrease the probability of selecting the model’s best choice.”

Gruber made an effort to but doesn't fully understand how SynthID works. LLMs select the next word randomly from a set probability distribution, so there is no "best choice" unless you run the LLM with a temperature of 0 which would give out terrible results. Anthropic runs a non-distorting version of SynthID that doesn't change the probabilities of the underlying distribution of tokens. It makes the watermark less likely to work over smaller samples but preserves text quality. I encourage the mathematically inclined to read the paper:

https://www.nature.com/articles/s41586-024-08025-4

For a set parameter size, if you benchmark an LLM on a single language, you'd get better results by training it on multiple languages rather than just the one. Scaling quality training data almost always increases performance as the neural network picks up general rules and an understanding of code that are language agnostic.

That's not how LLMs work. If you're talking about number of parameters, you wouldn't be able to reduce the size much by "removing" support for other languages.

Why add a middleman when you can ask your own account with /usage?

n=56 doesn't give you much information regarding the margin of error, unless you practice Tarot Reader's cat science. The standard deviation of outcomes and the difference between both outcomes matter just as much.

If I flip a coin 56 times and it always falls on head, I can be pretty much certain that it's not a fair coin. I wouldn't need to flip it 1000 times. We are all someone else's "know-nothing hack"...

Can we know for sure that the price drop is accompanied by a change in the model such as quantization?

On twitter, some people say that some models perform better at night when there is a less demand which allows them to serve a non-quantized model.

Since the models are only available through API and there is no test to check which version of the model is served, it's hard to know what we're buying...

It needs to not have economic value but it doesn't necessarily need to be useless and wasteful.

You can have growing deficits without actually "selling the nation" when you account for growth and inflation.

Say foreigners own 50% of US assets. When someone creates in the US a company ex nihilo and sells 10% of it to foreign investors, even though foreigners bought more US assets, this new wealth creation causes the % of foreign-owned US assets to diminish.

When foreign investors own US bonds, as inflation increases and the money supply increases, they end up owning less in real terms.

This doesn't account for investments that are limited in quantity such as real estate.

Also just because those are opposing forces, doesn't mean that foreign ownership wouldn't increase. As a matter of fact, when looking at the foreign ownership of US stocks, it has been steadily increasing.

Does this support the Selfish Gene theory of Richard Dawkins? They look like the smallest self-replicating molecules that he mentions in the beginning of the book.

Maybe you're polling the people that wanted to join before 2022? Otherwise, how would the jump in support in 2022 be explained?

Koyfin and yCharts do that. You're looking for "total return".

Wow this sounds like they trained it on ChatGPT outputs.

However I could not reproduce this output.

Relevant Matt Levine: "Why Would Anyone Buy Credit Default Swaps on the U.S.?"

https://archive.is/r0lI2

Thanks for explaining this!

It seemed to me that option #1 only applied if you were a fugitive and your whereabouts were unknown, not when you simply leave the state or country.

As to furtherance, I wonder how precise the definition of it is; is using proceeds of a crime, furtherance of a crime? Is obfuscating a crime, furtherance of a crime?

Edit: This answers most of my questions and the answer seems to be "it's complicated" https://sgp.fas.org/crs/misc/RL31253.pdf

The statute of limitation for wire fraud is 5 years. How did he get charged under that statute when he stole funds from Silk Road in 2012? Did they charge him for converting BCH into BTC in 2017?

I tried to look up the case but couldn't find anything that shed a light on that.

I tried playing blind chess against ChatGPT and it pretended it had a model of the chess board but it was all wrong.

As a third-worlder I welcome these jobs. They lower unemployment, raise average salaries, haggling power, increase the country's foreign currency reserves... Our minimum wage is similar to Kenya's and $1.46/hour after tax is way above the around 80c minimum.

It could make sense from an environmental perspective to produce and consume locally but for a job like this, the only reason to use US nationals would be to create local jobs.

None of the answers you received so far are correct.

The holding company of the foreign entity is in the US and can thus file for ch11. See who owns FTX Trading Ltd (Antigua) for example: https://d1e00ek4ebabms.cloudfront.net/production/8566b562-d5...

How can you say it's clear it wasn't cash? What's the source?

Also, FTX minted FTT out of nothing - effective cost zero - so no matter what they received in exchange, even if they had received nothing that is not an investment unless they received Alameda equity. I agree that lending is a form of investment but nothing says that they received a loan in exchange.

You could still be right, but it's all speculation :)

Nowhere does it say that FTX invested in Alameda.

Alameda invested in FTT which is minted by FTX which is not the same thing.

SBF said "We don't invest client assets (even in treasuries)". [0]

He then says the purpose of the transaction with Binance is to "clear out the liquidity crunches". [1]

How could there be a liquidity crunch if assets are not invested? You can't do a bank run on an entity that doesn't function as a bank and doesn't invest clients assets... Something is shifty.

[0] https://twitter.com/sbf_ftx/status/1589598285798707202

[1] https://twitter.com/sbf_ftx/status/1590012126701441025

I get a couple of gallons of olive oil in a small village directly from the press in November each year. It's not filtered so it has a lot of particles in suspension and it goes bad faster, maybe after a year or so. But it tastes nothing like supermarket extra-virgin olive oil, the taste is so much more concentrated.

Olive oil has a pretty short shelf life. The best way to get the good stuff is to make sure you buy:

- the latest harvest (they're usually harvested in early fall)

- non-filtered

- stored in a dark bottle, somewhere cool (light oxidizes it)

- single origin (if it's not, then they're likely mixing old rancid and new oils)

- first press, cold extracted

The fact that they closed his account frankly sounds like the outcome of an AML review.

Banks are deputized by the government to check for all sorts of suspicious transaction. They get fined when their procedures are not sufficiently stringent. They're not allowed to tip you as to why your account is being suspended or closed.

At any moment, any bank can lock your account for months without informing you of anything. It's unfortunately the power we decided to give them to prevent money laundering.

Fuel cost, depending on routes, number of business seats, seat occupancy is at about 25% of total costs [0]. 75% is what you call epsilon? Aircraft speed as barely budged since we transitioned from turboprop to jets. That's something you could call epsilon.

You're right that they start with business seats because they cost more. Business class seats cost 3-4x time economy while occupying less than 2x the space so the cost of fuel as a percentage of the ticket price is lower. It might not be obvious to you but I'm happy to explain it :)

[0] https://www.mckinsey.com/industries/travel-logistics-and-inf...

You're right.

That doesn't have much to do with the point I'm making. I'm saying if you double fuel usage, you don't double the cost of using the plane.

As cost of fuel is only a percentage of the price of the ticket, it's pretty obvious that there is a threshold as a percentage of total ticket cost under which spending 4x more in fuel to fly say 1.5x more passengers (because the airplane isn't flying 24/7) makes business sense.

That is obviously one of the reasons why they are starting with business class tickets because, fuel consists of a smaller percentage of ticket cost.

To everyone saying this will not work because it's more expensive... jets were more expensive than turboprops. And even though the former use more fuel, we prefer them.

The costs of flying an airplane isn't proportional to its fuel usage. The faster an aircraft is, the more flights it can perform per day.

The carbon emissions impact of flying a gas-guzzling supersonic aircraft aren't evident either. Of course, more gas is used per trip but fewer planes need to be manufactured. Since there is no supersonic business jet, it could also make sense for some people who used to fly private for the speed and convenience to reconsider as they may get faster to their destination by flying supersonic.

Well this article seems to be about the opposite of that, less precise than necessary or slightly outside of tolerance range!

This makes so little sense to me. I don't know how much information is made public by the Dutch justice system but I would really like to see an affidavit or something like it to understand what specific charges are levied against that developer.

I thought biomass was marginal but you actually produce almost as much electricity with biomass as with natural gas. I stand corrected. https://en.wikipedia.org/wiki/Electricity_sector_in_Germany