TaylorPhebillo

This is a reference to one the Recurse Center's social rules: https://www.recurse.com/social-rules

I was really impressed with how successful RC is at maintaining an environment where people can learn and grow. Part of that is certainly selection effects- the point of center is self directed growth around programming, and there's an interview process that I assume filters especially hostile people.

But I think the social rules do a lot too, and have been trying to pay attention to the effects on others when someone breaks them at work. No Feigned Surprise is a particularly important one around people who are trying to learn and already a little insecure. It's great when they've learned a new thing, and you want to celebrate that, not meet it with denigration!

How do prediction markets account for interest rates? I feel like I should be willing to pay no more than ~96 cents for a contract that will definitely resolve to a dollar in a year. Who puts up the other 4 cents?

I'm sincerely unclear- is the analysis here that smaller samples (fewer ballots processed by a tabulator) have a higher relative standard deviation than larger samples?

Occasionally but pretty rarely. By the time a trade has become actually risk free, as in "I can instantly buy X here and sell X there and make a profit", someone will have realized that and done the trade when it was almost risk free- the that the price is likely but not guaranteed to go up there- and made the trade first.

Almost always outside of crypto, the market makers and exchanges are different entities. Exchanges maintain order books- who is willing to buy or sell what, at what prices, plus a lot of rules about tie breaking, order visibility, "implied" prices (e.g. sometimes the combination of two products is logically equivalent to a third), etc. When orders "cross"- that is, someone is offering to buy at a price at least as good as someone is willing to sell for, the exchanges matches those participants and they are considered to have traded (though for a mix of technical and regulatory reasons, the trade actually settles two days later)

Market makers generally maintain offers to both buy and sell a product, generally ~all the time the market is open. For example, they might offer to buy up to 30 X for $0.99 or sell up to 70 for $1.01. If small buy and sell orders come in more or less randomly, the market maker will sell about as many X as they buy, for (1.01 - 0.99) a profit of 2c for each set of orders. The trick for a market maker is to offer the best price, so that they get any orders at all, while accounting for the risk that the person buying or selling from them (the liquidity taker) isn't just a random order, but is either market moving or correctly predicting the market is about to move- e.g. a market maker offering to buy a million shares of a X at $0.99 will lose a lot of money to someone who correctly predicted X is about to go to $0.70, and took them up on the full offer.

These are still fairly canonical, and what I got recommended when I started recently. Trading at the Speed of Light by MacKenzie is very good too- a lighter read, but more recent.

I see [1] speculate that Coinbase users could be considered unsecured general creditors, and therefore lower priority to repay than secured creditors. That does seem bad on it's own. But nothing about equity investors having higher priority on funds. Is there some extra subtlety around e.g. shares being convertible into debt or something weird?

[1] https://www.bloomberg.com/news/articles/2022-05-11/coinbase-...

Was this Amazon? Generally equity would replace the cash bonus in subsequent years. Compensation then becomes somewhat uncertain, but if it became a huge pay cut at that company and nowhere else, it seems expected that lots of employees would quit. That might be partially intentional.

[Deleted a previous reply out of an abundance of caution about proprietary strategies]

That matches my first order academic understanding of market makers, but not the more detailed work I've read, or the work of the QRs at the handful of market makers I've spoken to/worked with.

It didn't seem like connecting trades were a particular focus of the people I've spoken with- is that something that happens off exchange? My understanding is that would happen automatically on the exchange, and that primarily market makers connect trades across time, pocketing bid/ask spreads but accepting risk of adverse price movements while they hold that position. To mitigate that risk, they do a bunch of signal analysis to set the spreads low enough to be competitive, high enough to mitigate that risk. Some market makers famously pay and provide price improvement for small/uncorrelated orders from retail brokers. I was told that that was specifically because of the reduced risk of adverse market movement from retail orders.

Market makers are generally required to always provide buy and sell prices, or risk fines/losing their market maker status on an exchange, right? Those prices might be extremely unreasonable (e.g. in a flash crash setting sell prices to 9999999), but if there's a market maker, there is required to be a price at which they will execute a trade, whether or not they can hedge that risk or now.

That seems true, but in practice it seems like a market maker can't offer competitive spreads without having a decent sense of market direction, and they will have to take a position for at least a little while before closing out. So the line between a market maker and a trader who takes deliberate positions feels pretty fuzzy.

I'm not familiar with this, does this method require that each value is measured independently of the others? Because in that case, you'd want to use monthly inflation rates, rather than monthly estimates of inflation relative to 12 months prior, right?

I've played with it a little bit:

Copilot did pretty poorly when I tried using it with Julia- it kept suggesting Python code. I suspect it would do something similar in Elixir.

I'm also a vim person who doesn't want to use VS code, but I've gotten more than enough value to get into my first IDE (with vim keybindings). A lot of tedious C++ code is getting correctly auto-generated.

My hunch is that they aren't tailored toward ridiculous images exactly, but if they demonstrated "a woman sitting in a chair reading", it would be really hard to tell if the result was a small modification of an image in the training data. If they demonstrate "A snake made out of corn", I have less concern about the model having a very close training example.

ARM big little uses the same instruction set on each core, right? You wouldn't have one 8.2 core and one 8.4 core?

I hope so? But that's hard to study, and Givewell (in contrast to other effective altruism organizations) is fairly conservative about making claims and recommendations.

The enormous risk factor from the article is malnutrition. Givewell, an organization which specializes in estimating the cost-effectiveness of forms of charitable giving, suggests treating acute malnutrition could be extremely cost effective, costing a few thousand dollars to save a life.

https://blog.givewell.org/2021/11/19/malnutrition-treatment/

I'm not totally sure, but I suspect quantitate finance/market makers, of which tech is a big part, would do well during recessions- profits there seem correlated to volatility and volume of stock trading, which I'd guess would be high during a downturn and recovery?

That makes some sense. So buy-borrow-die with all the shares used as collateral generates no taxes, if you can find someone to lend for the full value of the shares. I wonder how high a ratio you can borrow if you're e.g. Musk or Bezos.

On the off chance anyone finds this comment, I found this page with the current modification proposals helpful as well: https://crsreports.congress.gov/product/pdf/IF/IF11812?fbcli...

I'm going through Understanding Software Dynamics by Richard Sites now, and it's the first book I've read that covers the practical performance implications of some of these new features, even if briefly.

I'm still learning about buy-borrow-die, but US estate taxes are higher than income taxes above ~$12 million, right? Much less capital gains. So buy-borrow-die would generate more taxes than buy-sell-spend?

I honestly can't tell how the Softbank vision fund is doing. News articles on it report record breaking profits and losses. The two vision funds were recently-ish reported as being worth $150b, but I can't tell what that translates to in annualized returns, or how that looks relative to the NASDAQ.

Is there meaningful, comprehensive data on anything like "What cities are focal points for technology work"? For example, number of software engineers, total software workforce compensation, number of funded startups, etc?

Edit: As a poor proxy, checking the most recent 200 levels.fyi entries for ", CA |" finds 37 entries when I checked, vs 12 for TX. Wikipedia says the San Francisco CSA has 9 million people, vs 2.2 million for the Austin MSA. There's lots of issues with that analysis (first, I think CA programmers are more concentrated than TX programmers), but to my surprise, it implies the the bay has more programmers, but Austin has more programmers per capita?

Do people have suggestions for better data, or better analyses of data?

https://www.levels.fyi/comp.html?track=Software%20Engineer

1/20 the cost sounds good, but can you post your math there, or why you think they can be beat by that margin? It sounds like a fun project to work on, but I worry that the price improvement estimate neglects a lot of what the cloud providers are charging for.

Agreed- I was really interested in the facts cited at the start, less so in the citation-less analysis later. So I looked into the report cited many, many times- it's from the Center for the Study of Partisanship and Ideology. Looking into them, influence watch says "The organization primarily reports on bias against conservatives. CSPI provides research grants of $2,000-$15,000 to applicants to study incidents or trends of bias against conservatives."

https://www.influencewatch.org/non-profit/center-for-study-o...

On a side note, I'd really prefer not stating something as fact, then linking only to a 200 page report without any hints about where to actually find the info.

Edit: I'm not thrilled with the linked report in general, but:

The author says

> Nearly a quarter of American social science and humanities academics support dismissing colleagues who have unorthodox views in areas such as immigration or gender differences.

From page 24 of the report, I see where the author finds this information, but it's worth noting the support for dismissing academics comes almost entirely from the topic "women and minorities lower organizational performance"- more conventionally conservative topics don't seem to have anything like a similar rate.

I can't find anything obviously like the 4 out of 5 PhD students discriminating against right wing scholars.

Do you mean as high as 3 seconds, or as low as 3 seconds? C++ compile times on a template heavy project I used recently were in the hours- you'd basically compile overnight and before you went to lunch.

I am poorly qualified to evaluate the claims of the paper, so as a proxy I looked into the qualifications of the authors. Dr Pantazatos is an assistant professor in the Psychiatry department at Columbia, with what appears to be a long record of papers in the field of brain imaging, with a shift this year (inferring from his twitter) to a focus on vaccine opposition preprints and petitions opposing vaccine mandates. Hervé Seligmann appears to be a currently unaffiliated geneticist, though he did (apparently falsely) claim to be affiliated with Karlsruhe Institute of Technology in a letter advocating hydroxychloroquine?

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8671645/

As always, I'll defer to real expert evaluation of the work, but my hunch is that the paper will be poorly received when that happens.

I used to be engaged in this part of the industry, but haven't looked in a while. So, sincere question: Are V100's still meaningfully popular? I know they were incredible for a long time, but I figured most usage would have shifted to A100 for high performance or T4 for cost.

I'm sympathetic to the view that "IBM is failing" lacks evidence, but trying to look at the evidence objectively, IBM doesn't seem to be doing well relative to the tech industry broadly. Their stock price is flat since 2000, while the SPY is up 5x. Their revenue is flat/decreasing since 2005 (as far back as I can find).

It's possible IBM never dies, or even never shrinks, but stays the same as the world grows around it. That seems like the same thing.

I'd be interested in learning more about that if you could share a name of the professor or project?

I think the Cortex-R chips are more targeted here- I think the V8-R chips are designed specifically for automotive use, so I assume they have appropriate ratings.