Astral Codex Ten

Gamestop

10 min read

Gamestop

Article

Gamestop is a recurring brand in the Astral Codex Ten archive, appearing 4 times across 4 issues between February 01, 2021 and December 20, 2022. The archive places it in contexts such as “you can still buy all the Gamestop stock you want”; “Gamestop stock price still above $100”; “Meme stocks like Gamestop and AMC sometimes remain mispriced indefinitely”. It most often appears alongside Ethereum, PredictIt, COVID.

Metadata

  • Category: Brands
  • Mention count: 4
  • Issue count: 4
  • First seen: February 01, 2021
  • Last seen: December 20, 2022

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Metaculus Monday
February 01, 2021 · Original source
But they're not quite the present. Right now I only know of three prediction markets, and none of them live up to their potential. As usual, it’s the government’s fault: betting on prediction markets is technically gambling, which makes it mostly illegal (of course, you can still buy all the Gamestop stock you want).
Mantic Monday: Predictions For 2021
April 26, 2021 · Original source
ECON/TECH 14. Gamestop stock price still above $100: 50% 15. Bitcoin above 100K: 40% 16. Ethereum above 5K: 50% 17. Ethereum above 0.05 BTC: 70% 18. Dow above 35K: 90% 19. ...above 37.5K: 70% 20. Unemployment above 5%: 40% 21. Google widely allows remote work, no questions asked: 20% 22. Starship reaches orbit: 60%
Predictions For 2022
February 01, 2022 · Original source
ECON/TECH 11. Gamestop stock price still above $100: 30% 12. Bitcoin above 100K: 20% 13. Ethereum above 5K: 20% 14. Ethereum above 0.05 BTC: 90% 15. Bored Ape floor price here below current price of $203K: 40% 16. Dow above 35K: 90% 17. ...above 37.5K: 40% 18. Inflation for the year below five percent: 90% 19. Unemployment below five percent: 50% 20. Google widely allows remote work, no questions asked: 50% 21. Starship reaches orbit: 90%
Inline links: here
Prediction Market FAQ
December 20, 2022 · Original source
Able to get top-1 strict accuracy of at least 90.0% on interview-level problems found in the APPS benchmark introduced by Dan Hendrycks, Steven Basart et al. Top-1 accuracy is distinguished, as in the paper, from top-k accuracy in which k outputs from the model are generated, and the best output is selected. By "unified" we mean that the system is integrated enough that it can, for example, explain its reasoning on a Q&A task, or verbally report its progress and identify objects during model assembly. (This is not really meant to be an additional capability of "introspection" so much as a provision that the system not simply be cobbled together as a set of sub-systems specialized to tasks like the above, but rather a single system applicable to many problems.) Resolution will come from any of three forms, whichever comes first: (1) direct demonstration of such a system achieving ALL of the above criteria, (2) confident credible statement by its developers that an existing system is able to satisfy these criteria, or (3) judgement by a majority vote in a special committee composed of the question author and two AI experts chosen in good faith by him, for the sole purpose of resolving this question. Resolution date will be the first date at which the system (subsequently judged to satisfy the criteria) and its capabilities are publicly described in a talk, press release, paper, or other report available to the general public. Even this isn’t perfect (which models are “the equivalent of” a 1:8 scale Ferrari 312?), but in practice once you get to this level of details people mostly stop worrying about this. Another method (mostly associated with Manifold) is to just leave it up to human judgment - specifically, the judgment of the person who made the market. For example, I could make a market in “By 2050, will there be an AI which Scott Alexander thinks qualifies as ‘human-level’?” This will force market participants to price in the risk that I have bad judgment or act dishonestly. But perhaps these risks are small. For example, I might say elsewhere what I think qualifies as “human-level” AI, or you might think human-level AI will be so obvious when it comes that I will definitely agree with you about it. As for honesty, this could be enforced either legally or by reputation. Someone who has resolved their past 100 prediction markets honestly will probably resolve this one honestly too, especially if they get paid to do so and will never get customers again if they lie. When we invest on the normal stock market, we trust that our brokers / the NYSE / etc won’t run off with our money, and this trust is usually well-deserved. Even when we make an online purchase, we trust that the store we’re sending our money to won’t steal it and refuse to send us the product. It would be an exaggeration to say that trust is a solved problem, but evidence from Manifold suggests that most people price in a <1% chance that well-known market makers with good reputation resolve dishonestly. If prediction markets got big enough, they could spawn trusted “resolution companies” who individual markets and market-makers could outsource their resolution to, for a fee. If these companies were ever dishonest, they would lose all their business from then on, so they would probably be as honest as other businesses like your broker / the NYSE / various online stores / etc. 4.7.1: Isn’t a lot of the “crisis of trust” around questions that might never have clear future answers? For example, consider the debate around whether Donald Trump is a Russian agent. Maybe no proof will ever come out either way. Or maybe some evidence will appear that seems to prove one side or the other, but people will continue to deny it for political reasons, and the problem of resolving the prediction market will be just as hard as the problem of answering the original question. Indeed, prediction markets aren’t very good at this, and are only fully trustworthy on questions where the true answer will eventually become apparent. Still, they might not be completely useless. For example, if you’re worried about Trump being a Russian agent because you expect him to pursue pro-Russia policies, you can start markets in whether he pursues those policies. Or you can start a conditional market (see 5.1) on whether, if Russia ever releases its past intelligence data many years from now, the data confirm/disconfirm that Trump was an agent. See Part 5 for other clever ways you might try to address this problem. 4.8: “Meme stocks” like Gamestop and AMC sometimes remain mispriced indefinitely. How do we know this won’t happen with prediction markets? Meme stocks are a type of Ponzi. It’s “reasonable” to buy Gamestop at some inflated price, because - who knows? - someone else might buy it at an even more inflated price tomorrow. And this can keep going arbitrarily long, or at least long enough for you to get out with a profit. Unlike meme stocks, prediction markets have a clear resolution date. If you’re predicting who will win the next election, the market will go to 100% or 0% after the election finishes. No matter how many memes there were, you wouldn’t buy a share in “the Democrats will win the election” for 99% the day before Election Day if you knew they would definitely lose. But that means prediction markets should be accurately priced the day before Election Day, which means you shouldn’t buy at an inaccurate price two days before Election Day, and so on. I can’t say for sure that no prediction market will ever get mispriced for meme reasons, but they should be much more robust against meme mispricings than the stock market. And even the stock market doesn’t have too many meme stocks. 4.9: How do prediction markets deal with outcomes in the far future? Suppose there is a question “who will win the 2100 election?” Currently it says 25% Democrats, 75% Republicans, and I believe it should be 50-50 (we’ll ignore third parties, or the possibility of America not existing in 2100, for now). So if I bet on the market, I can (in expectation) double my money. But there are many better ways to double your money by 2100. For example, if the stock market grows 4% per year, I should expect any money invested in the stock market to multiply by 20x in 2100. So just doubling it in a prediction market is a bad option. Realistically, this means prediction markets won’t work well for far-future events. These might be a better match for forecaster tournaments or some other structure, where we get the forecaster track records through present events, then use those track records weighting their far-future predictions (see also 5.5). There are already good forecasting tournaments on some far future events. But if you really wanted to use a prediction market, you could theoretically solve this by putting investors’ money in index funds while they waited. Then the winner would get their (and the losers’) original deposits and investment profits, and it would go back to being a better option than investing in index funds directly. In practice this seems complicated and I wouldn’t expect it to work. 4.9.1: What about predicting things that would make it impossible or pointless to win money, like human extinction? Again, these questions probably aren’t great matches for prediction markets, and you should use forecasting tournaments or some other method (see also 5.5). If you really wanted, you might be able to make it work in theory through a mechanism sort of like this one. 5. What are some clever uses for prediction markets? Here’s a non-exhaustive list: 5.1: Conditional prediction markets / decision markets Suppose the government is trying to decide whether to throw its weight behind Vaccine A or Vaccine B for some deadly disease. There are some experts behind both, both sets of experts accuse the other of being in the pay of pharmaceutical companies, and decision-makers don’t know who to trust. They might make two prediction markets, like: If we decide to go with Vaccine A, will at least X people die from the disease?
Inline links: Dan Hendrycks, Steven Basart et al, already good forecasting tournaments on some far future events, this one