antasvara

If you run a traditional short (borrow and sell the shares), you need to be right on the direction and the timing to make any money. One random surge in the price can kill your short even if said surge is short-lived.

GPU prices are sitting well above MSRP, even for older generation cards. Datacenter growth is also being pushed pretty hard by most of the cloud providers. More anecdotally, the model providers I use have been "throttling" my requests (usually by bumping them down to a lower model version) more frequently. Granted, that could be for profit reasons, but I think it all generally points to a lack of capacity.

>Is there a wait to get an inference provider or interruptions in availability?

I think that the inference providers and model vendors are mostly handling it behind the scenes with more aggressive quantization, increasing API prices, deprecating more generous monthly tiers, etc.

It's a few things I assume:

1. Inference providers are largely a commodity business. While demand outstrips supply, there are many providers of that supply, which does somewhat constrain how much leverage a provider has.

2. The models are not interchangeable as of yet (though this seems to be changing). A lot of people want the latest Anthropic or OpenAI model; as such, the labs still have considerable leverage.

3. A lot of the inference providers have used their leverage by doing equity deals with the labs, agreements to have some of the deal revenue spent on compute, etc.

Nathan Fielder is pretty well-known for his work in The Rehearsal and Nathan for You. Nathan for You resides in the docu-comedy genre, where you take advantage of real people's reactions to absurd situations for comedic effect. Other examples are things like Kitchen Nightmares (where Gordon Ramsey goes into real restaurants and "fixes" them). It's a well-done version of the idea, but not revolutionary IMO.

The Rehearsal, however, extends the idea further. In that show, he helps real people "rehearse" for hard conversations. It involves Fielder building extravagant sets and having absurdly detailed set-ups with multiple actors, then running the conversation multiple times. What makes this "different" is that now the real person is partially in on it? Like, they know that they're on a set. They know that this is a simulation. So the show is more "meta" than previous entries in the genre, opening up other avenues for humour.

I say this because You Can See Everything is best understood as a further extension of those projects. Even the name indicates this, because by nature of a documentary, you can't see everything. You have a real person in Elizabeth Holmes, being interviewed by a person in Fielder that's known for his stunts taking advantage of "real people," where he watches fictionalized versions of Holmes with Holmes herself? Viewed in light of his previous projects, this feels like an extension of them.

But even with that, I think the setup does allow you to see something real in Holmes. She also knows how people "want" to see her! So there's also a sort of meta-narrative there too, where you know Holmes is on her best behavior to try and dispel the narrative from those fictionalized TV characterizations. The fact that she comes off poorly in the documentary says something, even accounting for the fact that Fielder is trying to sell a narrative.

It's a real mind game, but I think it's representative of Holmes because of that. She is going to jail for a fraud that amounts to a similar sort of mind game, she agrees to a documentary to show herself differently, but that documentary is also a mind game. Wild stuff.

>You can't 'blame' the AI, as it can't be held accountable.

Seems like more of a semantic distinction IMO. Yes, I can't technically "blame" the AI because it's not accountable. But what word would you suggest for "I had [X technology] handle [Y task] and it failed to perform that task?" I can't "blame" my router if I lose internet and it prevents me from jumping on a Zoom call, but it's also true that the router failed to do the thing it was supposed to do.

I don't know, it feels like "reading a claims summary and classifying the type of claim" should be the bread and butter LLM use case? Not that executives are blameless for pushing AI everywhere, but we should also be able to "blame" AI for doing poorly at a task it's supposed to be good at.

An exclusive theatrical window is one of the big reasons to go to a movie theater: you want to see a movie before it's available on streaming. Removing copyright seems like it would harm theaters pretty significantly.

I think the point is more "why is the outcome of an event like an election considered a commodity" and not a question of the value of event contracts?

Like, why would the same agency regulating wheat futures also be responsible for regulating event contracts? I know the simple answer is that Congress said they should be, but conceptually it's a bit odd.

It's humans from other banks/funds bidding on the block of stock. As far as timing, for this situation it's basically overnight for regulatory and price reasons. Regulatory because there are legal margin requirements for levered positions and you can't handle the price going much lower, and price because if you had to sell this on the open market you'd keep selling shares for less and less.

So JPMorgan/Goldman prepare all the info on the book and start calling institutional investors after the market closes. The funds and banks prepare bids, there's some negotiation, and the block is finalized before trading opens the next day.

Speed does matter, but you're only calling investors you know "can" close the deal (i.e. they'll have enough capital to buy it all that day/night). So it's more of a price question than a speed one at that point?

And really, nobody wants the downward spiral of a fire sale in the tech sector. Someone will make money on that chaos, but it's a lot of risk when you can lock in a discount with the block trade.

They might not have known it was SA specifically. That being said, they definitely knew a large fund with leverage was buying these stocks. The mechanism here (and I'm not an expert) is:

1. SA wants to buy stock with leverage. You do that through a major bank via total return swaps. Essentially, SA pays X% of the value on $100 of stock (for 4x leverage you'd pay $25) plus an ongoing financing fee (call it 5% a year), then you get the return/loss on that $100 of stock. SA was in these agreements with JPMorgan and Goldman Sachs.

2. The bank, because they don't want to actually hold that risk, goes out and buys $100 of stock.

3. Citadel and others see JPMorgan buying lots and lots of this stock. That's confusing, because normally JPMorgan wouldn't be making a huge directional bet on a stock. They deduce that a large fund is buying the stock.

4. Citadel starts widening their spread (the difference between what they'll buy a stock for and what they'll sell it for). They hedge some of this as best they can, or temporarily live with the risk.

5. SA, the highly leveraged fund buying volatile stocks, inevitably blows up because volatile stocks swing around in price. A dip causes margin calls.

6. JPMorgan or Goldman need to sell the stock fast, because SA is close to dipping below their required margin (i.e. SA paid $25 for $100 in stock exposure, the stock drops to $90, JPMorgan asks for more money because the stock went down by too much).

7. Citadel offers to buy all of the stock from JPMorgan. Because they're doing it in one big block, JPMorgan doesn't lose money selling on the open market (once you start selling, each successive sale is for less money because there are more people selling than buying). Citadel is compensated for this by getting a discount to the asset value (the stock is worth $90, Citadel gets to buy it for $81).

So Citadel didn't do anything to "set up" SA. But because they're hyper-aware of market dynamics, they would have known that someone is going to need to sell stock if the market takes a turn on these names.

Even in the optimistic case where SA did have alpha, the position sizing was way out of whack. Based on the volatility of the stocks they were buying, the Kelly Criterion meant you'd need to expect a 900% annual return on the stock before leverage to justify being 4x levered.

What guys like Leopold either don't understand or understand but ignore is that being right directionally and being right on market timing are two different skillsets. When you've juiced a stock by 800%, the existence of alpha pales in comparison to your vulnerability to the stock market.

Just to jump in: Citadel buying this portfolio says nothing about how Citadel feels about the stocks. It's the bread and butter of large HFT hedge funds; if you see someone that has to sell stock, you leverage the fact that you can buy all of it to get a discount versus the asset value. Reports are saying that Citadel was able to buy the portfolio for ~10% under the market value, all at once. That's a no-brainer because you both get a discount and avoid driving the price down by buying small pieces over the course of a week.;

If I were a betting man, I'd bet that Citadel was also selling to Situation Awareness while they were on the way up. At some point, SA had juiced their stock prices so much that no "rational" investors (those that have a view of the stock based on some amount of fundamentals) would be on the other side of the trade. It's retail investors, bandwagon investors, and Citadel-caliber funds. This situation (over-leveraged company blows up due to some volatility) happens all the time in commodities trading, which is where Citadel started.

Sure, a college student can tell you if a company is making or losing money. But there are ample examples of "good" companies crashing because they didn't react to shifts in the market, which requires projecting into the future.

Blockbuster was good until it wasn't. Sears was good until it wasn't. Barnes and Noble was good, then it fell apart, and now it's good again. The lesson to learn is that by the time your financials say things are taking a turn, you're a year to multiple years late to start fixing the problem.