Category: Web/Tech
Redux, my CWT with SBF
I won’t indent:
COWEN: I have some crypto questions for you. Is there, in fact, any way to coherently regulate stablecoins? I see what the proposals say: It’s all about capital requirements, deposit insurance, treat it as a bank account, like a new kind of money market fund. Can that possibly work? Doesn’t it end up having to be applied to all of crypto, all payments companies, PayPal, whatever else? What’s really there that they can do?
BANKMAN-FRIED: That’s a really interesting question. First of all, I will say, I think there is something that does work compared to the current environment, but I’ll get to your point — it’s actually a good one. If you just said, “Look, all stablecoins have to be fully backed by the dollar and have to have audits to confirm that they are in a bank account,” that would get a pretty safe product that was well understood, well regulated, and frankly would be, from a product perspective, just as good as current stablecoins. That’s what all the stablecoins are doing today.
It’s a mess because there’s no clear regulatory framework for them to fit into and to have oversight of that. So, part of my answer there is, basically, yes, I think that framework would solve the current problems that people have in a pretty clean way. But you have a good point there, which is, how about PayPal? There are all these things that we don’t call stablecoins right now, that we call something else.
COWEN: PayPal promises me a dollar, and they give it to me. I’m happy, right?
BANKMAN-FRIED: Exactly. In fact, a lot of these look a lot like stablecoins when you drill into it. When you really dig into it, what is the difference between PayPal and USDC? I guess there’s some differences, but I think there are more similarities than there are differences, to be honest. What does that imply for PayPal? You can just say, whatever it implies, stablecoin is not PayPal — it’s how it is.
I think there would be a big improvement over the current world, where it’s the same thing but without regulatory oversight and with a lot of random drama because of this. But I do think that it gets to this question of, “Wait, but banks are allowed to rehypothecate dollars. Banks are allowed to do all manner of wacky things with their deposits.” Are stablecoin companies allowed to? If not, is it obvious they shouldn’t be allowed to? And how should that be governed and regulated?
Maybe the answer is, whatever: The banks will do that on their behalf, the banks where they hold their assets, and then pay them interest for the right to do that — although, of course, right now banks aren’t actually paying interest, really . . .
COWEN: Not to me. Also, how stable does a stablecoin have to be to be regulated as such? If there’s any regulatory definition, won’t a lot of people just camp their crypto assets to be just slightly more volatile than wherever the line is drawn, or you’d just end up regulating all of crypto? How does that work?
BANKMAN-FRIED: This could go in a few ways. Is your thought that people will attempt to get just barely into the regulatory system or just barely out of it?
COWEN: Maybe both, but a lot of people will go out of it. So I’ll issue something. I’ll call it a “not stablecoin,” but de facto, it will be very stable. But also, “Oh, it’s just sort of an accident. Oh, who knows what the markets going to do today?” It’s just stable for decades. How do you regulate that?
BANKMAN-FRIED: Oh, that’s a really good question. Of course, what it gets you is this question of, what if a stablecoin didn’t promise to be a stablecoin? Is it bad that it’s backed by the dollar? Does it somehow make it worse from a regulatory perspective? Why is it being held to a higher standard?
COWEN: Exactly.
BANKMAN-FRIED: I do think there’s a little bit of an answer here, although I also think that this is getting at another point, which is, you could reasonably say, “Look, are consumers doing what they’re doing with their eyes wide open?” If there’s sufficient disclosures and transparency, shouldn’t people be allowed to use stablecoins with some risk in them?
I think that’d be a reasonable thing to think, but if you put that aside for a second, you say, “No, absolutely not.” Here’s one difference between that and the stablecoin, which I think is relevant, is that a stablecoin is not just stable in one direction. It’s stable in both directions. In particular, if you’re an investor, and you buy a stablecoin, you have downside risk but not upside. If somehow the stablecoin company makes money, you’re probably not going to get any of that, but if it loses money, somehow, you’re probably on the hook for that.
So, there is something a little asymmetric going on here for the consumer. I think it wouldn’t be crazy from that perspective to think that there should be some protection here and that maybe there should be regulation if consumers are only given one side of exposure, but I don’t think that’s obviously true. I think you’re making a decent point.
COWEN: Now, if we look at DeFi, there are some forms of obvious, explicit leverage, like people borrow money to participate in the system. But those aside, I’ve learned over my life, if you look at any system, any institution, typically there are forms of hidden implicit leverage in those institutions. Might be good, might be bad, but it’s there, and in a sense, you don’t understand the institution until you understand where’s the implicit leverage in this game. In DeFi, where is the implicit leverage? Is it rehypothecation, or where is it? What is it?
BANKMAN-FRIED: Ignoring the explicit leverage of borrow-lending protocols —
COWEN: Yes, which is easy to see, right?
BANKMAN-FRIED: Which is easy to see, yes. So, what happened in 2008? What caused the collapse in a lot of things? That’s sort of a dumb question, but one of the things that led to this is that no one knew how much leverage there was, really, in the system. As you said, there’s always implicit leverage, and in this case, it was all of these bespoke OTC swaps between banks that basically didn’t get reported anywhere. In fact, those got rehypothecated again and again and again. No one was keeping track of the total notional fees. It was impossible to — they weren’t public.
One thing you could do is look for a similar thing in crypto. You could look for OTC transactions. You could look for OTC swaps that live on. You could look at OTC borrow-lending. Those are in crypto. Are they in DeFi? It’s sort of ambiguous — they touch all areas of the crypto ecosystem.
But that’s an area where I think there’s some dubiously accounted-for leverage. I think that’s one answer to that question. Where else is there leverage that sort of is implicit? Rehypothecation sometimes, although in DeFi, because it’s all on-chain, it has to be pretty explicit if it’s going to be rehypothecated, but you’re not . . .
COWEN: But it’s hard to see, right? If you traced everything, you could find it, but no one’s actually watching it. Or are they?
BANKMAN-FRIED: Well, they’re halfheartedly watching it maybe, is how I’d put it, which is not great. Maybe full-heartedly watching it. I could imagine arguing for people full-heartedly watching it, and that would be a reasonable thing for them to be arguing for. In particular, if someone releases a protocol, there’s a question of, well, is that protocol rehypothecating? You just look at the code and see if it can rehypothecate, right?
In general, people actually often do know whether each protocol individually can rehypothecate, which is a separate question from whether they, as a group, can or whether they are or something. But in fact, most of these aren’t. Most DeFi protocols are not doing things beyond what they literally say they’re doing, and so the amount of leverage they introduce into the system mostly is what they say they are.
But here are some hidden things. First of all, you take one leverage thing, you put in another leverage thing, so DAI. DAI is an algorithmic stablecoin. Like other algorithmic stablecoins, it is not perfectly stable. It’s not perfectly stable because it’s not backed by the US dollar. It’s backed by crypto assets that could have price movements. It’s very overcollateralized. DAI can then be used as collateral on some borrow-lending protocols in crypto. That’s one form of rehypothecation in DeFi markets that you can trace through. It is, in theory, public, but it’s not super easy, necessarily, to trace through.
COWEN: Now, for mathematical finance, as you know, we at least pretend we can rationally price equities and bonds. People started with CAPM. It’s much more complicated than that now. But based on similar kinds of ideas — ultimately arbitrage, right? — if you think of crypto assets, do we even have a pretense that we have a rational theory of how they’re priced?
BANKMAN-FRIED: With a few of them, not with most. In particular, let’s talk about Dogecoin for a second, which I think is the purest of a type of coin, of the meme coin. I think the whole thing with Dogecoin is that it does away with that pretense. There is no sense in which any reasonable person could look at Dogecoin and be like, “Yes, discounted cash flow.” I think that there’s something bizarre and wacky and dangerous, but also powerful about that, about getting rid of the pretense.
I think that’s one example of a place where there is no pretense anymore that there is any real sense of how do you price this thing other than supply and demand, like memes versus — I don’t know — anti-memes? I think that more generally, though, that’s happened to a lot of assets. It’s just less explicit in a lot of them.
What is Elon Musk’s greatest product ever, or what’s his most successful product ever? I don’t think it’s an electric car. I don’t think it’s a rocket ship. I think one product of his has outperformed all of his other products in demand, and that’s TSLA, the ticker. That is his masterpiece. How is that priced? I don’t know, it’s worth Tesla. It’s a product people want, Tesla stock.
COWEN: But the prevalence of memes, Dogecoin, your point about Musk — which I would all accept — does that then make you go back and revisit how everything else is priced? The stuff that was supposed to be more rational in the first place — is that actually now quite general, and you’ve seen it through crypto? Or not?
BANKMAN-FRIED: Absolutely. It absolutely forces you to go back and say, “Well, okay, that’s how cryptocurrencies are priced. Is it really just crypto that’s priced that way?” Or maybe, are there other asset classes that may claim to have some pricing, or purport to, or people may often assume it does, but which in practice is not exactly that? I think the answer to that is a pretty straightforward yes.
It’s a pretty straightforward answer that you look at Tesla, you look at a lot of stocks right now, you think about what determines their market cap — the discounted cash flow? Yeah, sort of, that plays a role in it. That’s 30 percent of the answer. It’s when we look at the meme stocks and the meme coins that we feel like we can see the answer for ourselves for the first time, but it was always there in the other stocks as well, and social media has been amplifying this all over the place.
COWEN: Is this a new account of how your background as a gamer with memes has made you the appropriate person for pricing and arbitrage in crypto?
BANKMAN-FRIED: Yeah, there’s probably some truth to that. [laughs]
Here is the full dialogue.
The USG program with the highest benefit-cost ratio?
I am talking here about individual, discrete activities of the government, not general or overall functions. How about this for a nomination, namely US Embassy air quality tweets?:
Over 4 million premature deaths per year are attributed to air pollution, most of which are in low- and middle-income countries where residents do not have access to reliable information on air quality. We evaluate a large-scale program that provided real-time air-quality updates at over 40 US diplomatic sites around the world with poor preexisting monitoring. We find that the embassy monitoring program led to substantial reductions in fine particulate concentration levels, resulting in substantial decreases in the premature mortality risk faced by the over 300 million people living in cities home to a US embassy monitor. Our research indicates that monitoring and information interventions that draw attention to poor air quality in developing countries can generate substantial benefits.
Did you take these benefits into account last time you were bitching about Twitter? Probably not.
Via the excellent Kevin Lewis.
EDS
If I write about a government program in the abstract, some of you respond in a pretty reasonable manner. Instead, if I tie the program to the status of a well-known personality, such as Hillary Clinton, Obama, Biden, Trump, and so on, the quality of the responses is much lower. Including from very smart people.
Take this insight to heart and apply it to your current thoughts about the new Twitter.
How many of you have written me to say that people “won’t pay $8 a month for Blue Check,” or whatever the latest suggested price might be?
Note from Elon’s own words that a) “You will also get: – Priority in replies, mentions & search, which is essential to defeat spam/scam – Ability to post long video & audio – Half as many ads”, and b) “And paywall bypass for publishers willing to work with us”, and c) “This will also give Twitter a revenue stream to reward content creators”.
Please do read those words carefully.
Now I do not myself pretend to know what will work for Twitter. But one implication of this proposal is that a free version of Twitter still will be available. Is it so crazy to think that the forthcoming free version of Twitter will be “good enough” to keep the current users on?
Note also that under the proposed new regime, payments go both ways! Hardly any of the critics note this.
Or do you think markets are most efficient when all payments are set to zero and kept there? Maybe in some settings, but overall? I just do not see why this kind of plan is so doomed to fail. Let’s pay the creators who attract other Twitter users to create more and to induce more Twitter impressions. An externality is present, right?
As a friend of mine once said “Never underestimate Elon”…of course I would invoke more rational responses if I instead wrote “Do not underestimate the current Twitter management team and proposal.” But I won’t because I, like Elon, sometimes enjoy trolling you all.
Elon’s current Twitter strategy
Not macro, but super-micro. Relative status is what gets people talking, and what is more relative status than “Blue Check” on Twitter. And so everyone is talking about Twitter over the last few days.
Duh.
Here is the latest pricing proposal:
Twitter’s current lords & peasants system for who has or doesn’t have a blue checkmark is bullshit.
Power to the people! Blue for $8/month.
— Elon Musk (@elonmusk) November 1, 2022
And adjusted for PPP. Read the whole thread, a lot of privileges will come with the status, and I suspect spam problems will de facto force most people into this option. See you there! And Elon is already ahead of the critics on this one, and was all along.
The MR bot, John McDonnell working with Bryan Gilbert Davis
From John:
“I actually built a bot to produce “an MR take”:
It’s live here: https://www.vibecheck.network/
Intro video here: https://www.loom.com/share/e307765429db4fe38efd2fc822bb4529
(We applied to the stability AI grant but didn’t get it)
I think the results need to get a bit better (“Are aliens real” is pretty good, “What should I eat in Oaxaca?” is not good). I put some decent queries in the postscript.
Regarding your article I actually think attribution will be really important. AIs have a tendency to confabulate (the term d’art in the literature seems to be “hallucination”). Attribution is what tells you that the information is real and wasn’t just invented or pulled from some other hallucinating AI.
What would you want this world to be like? A couple thoughts I had:
- I agree that compactness will be desirable, but also “authentic” non-AI content will be valued. Imagine custom briefings about daily trends with a few of the “best” tweets pulled into the briefing.
- Who do you trust? You don’t want your AI pulling from any old source. There are few universal authorities anymore. I’m imagining that there may be “webs of trust” where e.g. I trust Tyler, he trusts someone else so I trust that other source at least a little. The DSA might publish lists of “approved socialist sources”.
- How do you want to feel? Can I make you a feed that’s “uplifting”? “Insightful”? Maybe if people can control their own algorithms we get a sort of do-over on social media and people get control of their attention back.
- “Idea lineage” … can I select the term “mood affiliation” on a recent post and get a lineage of the way the blog uses that term, hopefully get a contextual definition, etc…
- I’m curious about combining every source in someone’s digital life. What if we could link article, highlights, notes, emails, Slacks, etc? This is sort of huge but could get interesting.
If we were to iterate on Vibecheck are there ideas you’d like to see prototyped?
…PS a few queries that have decent answers:
* What is the Great Stagnation?* What do Russians believe about Ukraine?* Are aliens real?* Should I start a startup?* Should I support the TPP?”
TC again: I am already impressed. But I think many of you don’t understand what this looks like when hundreds of trillion (yes, trillion) of parameters are brought to bear on the problem, as will someday (soon) be evident.
PayPal boo
They have added the objectionable clause and fine back into their terms of service. It is important to respond appropriately to this.
How AI will change everything on the internet
That is the topic of my latest Bloomberg column, Washington Post reprint here, and yes people this is for real. Here is one excerpt:
Change is coming. Consider Twitter, which I use each morning to gather information about the world. Less than two years from now, maybe I will speak into my computer, outline my topics of interest, and somebody’s version of AI will spit back to me a kind of Twitter remix, in a readable format and tailored to my needs.
The AI also will be not only responsive but active. Maybe it will tell me, “Today you really do need to read about Russia and changes in the UK government.” Or I might say, “More serendipity today, please,” and that wish would be granted.
I also could ask, “What are my friends up to?” and I would receive a useful digest of web and social media services. Or I could ask the AI for content in a variety of foreign languages, all impeccably translated. Very often you won’t use Google, you will just ask your question to the AI and receive an answer, in audio form for your commute if you like. If your friends were especially interested in some video clips or passages from news stories, those might be more likely to be sent to you.
In short, many of the current core internet services will be intermediated by AI. This will create a fundamentally new kind of user experience.
It is unlikely that the underlying services will vanish. People will still Google things, and people will still read and write on their Facebook pages. But more will move directly to the AI aggregator. This dynamic is already happening: When was the last time you asked Google for directions? They exist online, of course, but if you’re like me, you just use Google maps and GPS directly. You have in effect moved to the information aggregator.
Or consider blogs, which arguably peaked between 2001 and 2012. Then Twitter and Facebook became aggregators of blog content. Blogs are still numerous, but many people get access to them directly through aggregators. Now that process is going to take another step — because the current aggregators will themselves be aggregated and organized, by super-smart forms of machine intelligence.
The world of ideas will be turned upside down. Many public intellectuals excel at promoting themselves on Twitter and other social media, and those opportunities may diminish. There will be a new skill — promoting oneself to the AI — of a still unknown nature.
Of course there is more at the links above. I could have written a much longer column of course. Just imagine asking the service of your choice for “a Tyler take” or “an Alex take.” Solve for the whole equilibrium! Many more institutions are aggregators than you might at first think…
Nat Friedman and Alexandr Wang discuss AI
Emergent Ventures, 22nd cohort
Emily Karlzen, Arizona, Founder and CEO of Arch Rift, to develop an astronaut helmet for commercial space flight.
Mehran Jalali, for building energy storage systems, NYC, grew up in Iran.
Kyle Redlinghuys, a further award, recently launched an API to make the data from the James Webb Space Telescope available.
Pranav Myana, 18, University of Texas, Austin, working on incorporating renewable power into the grid.
Brian Chau, Waterloo (Canada), general career support for writing and podcasting. Here is his Substack.
Cathal J. Nolan, historian, Boston University, to write a book on the relationship between war and progress. Just learned he was born in Dublin.
Cynthia Haven, Stanford University, to write a book on John Milton and the 17th century. Twitter here.
Harsehaj Dhami, 17, lives in Ontario, to visit a Longevity conference in Copenhagen. LinkedIn here.
Jackson Oswalt, Knoxville, builds things, AR/XR stuff, for general career support. In the Guinness Book of World Records for achieving a nuclear fusion reaction at age 12.
Miguel Ignacio Solano and Maria Elena Solano, Bogota/Cambridge, MA, co-founders of VMind, an artificial intelligence project.
Brian Kelleher, 18, Dublin, to improve software for doctors.
Devon Zuegel, to develop a new village and community, Twitter here.
Rodolfo Herrera, Pensamiento Libre, market-oriented Facebook and YouTube videos for Mexico.
Alia Abbas, 19, Maryland, to study biochemistry and materials and for general career development.
There are two other projects not yet ready for public announcement.
Ukraine tranche: There is now a new Emergent Ventures Ukraine.
Julia Brodsky, Maryland, former instructor of astronauts. To support educational efforts to teach on-line STEM and other subjects to Ukrainian children in refugee camps.
Uliana Ronska, 17, Prague and Netherlands currently. She is doing research on problems of triangulating fast-moving stars. It was also under her leadership that her team won ExPhO, CETO, and 2 all-Ukrainian Motion physics olympiads. For general career development.
Demian Zhelyabovskyy, currently at Bromsgrove School in the UK, from Kyiv. Last year he won first place in the All-Ukrainian Physics Marathon; also he and his teammates won the Experimental Physics Olympiad (ExPhO) and Computer Experiment Team Olympiad (CETO). For general career support, and for the physics paper he is currently co-authoring.
Tymofiy Mylovanov, representing the Kyiv School of Economics, to nurture talent development for Ukraine. Tymofiy as an individual was also the very first Emergent Ventures winner.
Congratulations to all, I am honored to be working with you!
How much will AI succeed in the arts?
That is the topic of my latest Bloomberg column, and let us start with what is likely to work well:
It almost goes without saying that the AI revolution currently underway is impressive. It is likely to have a huge impact in some parts of art world, such as the commercial sphere — consumers are generally not interested in who made any given ad or logo. It either works or it does not, and those conditions favor the machine. AI will also give the world quality (automated) personal assistants and autonomous vehicles, among many other advances.
But here is the problem:
Consider music. If Taylor Swift’s or Beyonce’s songs had been made by a software program, with no star at the microphone, would they be nearly as popular? It is no accident that Taylor Swift has more than 227 million Instagram followers — her fans want more than just the music, and that extra something (at least so far) has to be supplied by a living, breathing human being.
In the world of the visual arts, too, collectors are often buying the story as much as the artist. Even the experts have trouble distinguishing a real Kasimir Malevich painting from a fake (he painted abstract black squares on a white background, with a minimum of detail). The same image and physical item, when connected to the actual hand of the artist, is worth millions — but if shown to be a fake, it counts for zero.
Here is a qualification:
There will undoubtedly be many collaborations between AI and human creators, with the humans put forward as the public face of the joint effort. Periodic scandals about authorship will surface (“did he write any of that song?”), just as allegations of cheating with AI have risen to prominence in chess. AI-generated art will attract the most interest when the aesthetic of the creation and the personality of the human accompanist appear to be in sync.
And this:
Imagine that you took some souped-up future version of GPT-3 and fed it all the world’s texts through the year 1500. Would you expect it to be able to come up with something equally important and original as Shakespeare’s plays, or Newton’s Three Laws? How about Strawberry Fields Forever? Skepticism on this point has hardly been refuted by recent advances, however impressive they may be.
We watch Magnus Carlsen, not Stockfish vs. Alpha Zero, even though the latter match is of higher quality and likely more exciting too, at least in terms of moves over the board.
PayPal remains out to lunch
That was then, this is now
On May 9, PayPal is participating in the Net Neutrality Day of Action through its trade association Internet Association to support a free and open internet…Throughout the day of May 9, IA and all of its member companies will communicate their support for enforceable principles such as a ban on blocking, throttling, and paid prioritization…Those lack of choices can lead to Internet Service Providers (ISPs) playing gatekeeper to content and access.
Net neutrality for thee, but not for me…
The $2500 fine (read it and weep)
At first I thought this was Twitter b.s., but no I have been referred to the PayPal website update:
You may not use the PayPal service for activities that…involve the sending, posting, or publication of any messages, content, or materials that, in PayPal’s sole discretion, (a) are harmful, obscene, harassing, or objectionable, (b) depict or appear to depict nudity, sexual or other intimate activities, (c) depict or promote illegal drug use, (d) depict or promote violence, criminal activity, cruelty, or self-harm (e) depict, promote, or incite hatred or discrimination of protected groups or of individuals or groups based on protected characteristics (e.g. race, religion, gender or gender identity, sexual orientation, etc.) (f) present a risk to user safety or wellbeing, (g) are fraudulent, promote misinformation, or are unlawful, (h) infringe the privacy, intellectual property rights, or other proprietary rights of any party, or (i) are otherwise unfit for publication.
You can’t buy Traveler’s Checks (with PayPal) either! For each offense, at the discretion of PayPal: “Violation of this Acceptable Use Policy constitutes a violation of the PayPal User Agreement and may subject you to damages, including liquidated damages of $2,500.00 U.S. dollars per violation, which may be debited directly from your PayPal account(s) as outlined in the User Agreement…”
Update: The policy partially has been walked back, I don’t yet know the full details. And note that in 2018 PayPal was supporting net neutrality!
“But are you long volatility?”
As many of you know, when I confront mega-pessimists I like to ask them “But are you short the market?” Not once in my life have I heard a satisfactory rejoinder to this query. (The last answer I heard was “I will be!”.)
I now have a new question for those who see a reasonably high likelihood of AGI. Of course AGI could wreck the world, make the world a whole lot better, or simply overturn multiple sectors of the economy, in both good and bad ways. If you believe in AGI, typically you believe it will matter a lot, though there is considerable disagreement on the cost-benefit ratio.
So basically you should go long volatility, both of the aggregate market and sector-by-sector.
Can I call them AGIers? (“Aggies” is already taken.) AGIers ought to be long volatility.
Are they? I have started asking this question, and I plan to continue the practice.
India fact of the day
Not even 200 of the approximate 10,000 students from the Indian Institutes of Technology took up positions outside India last year. Fifty students, who make up the largest contingent, will be leaving from IIT-Bombay, followed by 40 from Delhi, 25 from Kharagpur, 19 from Kanpur, 13 from Madras, 17 from Roorkee and five from Guwahati. In 2012, 84 IIT-B candidates had accepted international job offers.
“Compared to 20 years ago, a very small percentage of students go abroad today. This is contrary to the general perception ,” says IIT-Delhi director V Ramgopal Rao. “Twenty years ago, 80% of the BTech class used to go abroad. Now these numbers are insignificant.”
Here is the full story, from 2017, via Elad Gil.