AI-Based Long-Tail Asset Pricing Models with Upshot HQ

Transcript

[Music] what's going on guys welcome back to the rollup today we are talking with a very special guest Nick Emmons who is a contributor to the project at upshot and upshot is a combination of AI and blockchain to uh loop in as many buzzword as we can here uh we're going to be talking about the intersection there of how they Leverage machine learning algorithms uh to provide a better defi experience we're going to be talking about longtail assets um and what they bring to the uh to the ecosystem so Nick uh thanks for being here uh before we uh dive into some of

the material Andy how you doing today yeah all right I'm a longtail asset Maxi today so I'm excited to uh to talk more about how we can get as D gen as possible these days welcome to the bull market Nick thank you thank you it's our pleasure um so so uh your your your background before jumping into uh upshot was it was it h crypto related or or did you kind of jump in here through a different avenue yeah I've been in crypto in in various capacities since like 2014 is 2015 is um prior immediately prior to upshot I was leading blockchain at John Hancock and

Manu life couple very large asset managers insurance companies um and we were doing a lot of research around like uh the pricing and pooling of of longtail non-parametric risk um and kind of building kind of decentralized insurance systems um that kind of catered to these like more esoteric longtail risks um and that that lent itself kind of pretty well and aligned pretty well with the kind of direction that we we started upshot in in kind of working through pricing longtail assets and and longtail markets yeah nice like my understanding of non parametric is that does that mean it's like almost it doesn't adhere to a certain like valuation model like

there's no single parameter you can trace that that links to the the value of the the cryptocurrency sort of yeah it's it's around sort of the like outcome and and the definition of the outcome so an example of parametric risk is like I'm taking some some uh crop insurance policy out and I'm going to get paid if like a median kind of temperature is is above this there's like a specific parameter that can autonomously close out the insurance contract in One Direction non-parametric is um uh things like health insurance or or types of insurance where there's some sort of like subjective nature in determining what can close out the the contract which is obviously like the the harder

kind of sector of insurance policies to bring into uh a crypto or decentralized setting cool cool so before we we get into like some of the longer tale asset like models that you guys have produced um we we've had Nexus Mutual um and a couple other decentralized insurance providers on on the channel could you kind of like explain like maybe you've you've taken a look at those guys through the research that you did with John Hancock what are some of the similarities and differences by looking at it through the the like centralized John Hancock lens and then looking at it through the the decentralized uh like Nexus Mutual L yeah I think what's what's like so compelling about

decentralized insurance is it removes a lot of the these like uh administrative costs or or expenses related to just the bureaucracy of running a large insurance company um I I think this this is a fluctuating number I think close to half of the the costs related to Insurance are administrative it's just because you have this incredibly large kind of inefficient bureaucratic entity in the middle of all of these core functions of insurance that's adding a bunch of bloat and every dollar you add to overhead is a is a Dollar Plus added to the minimum Market size of like what is viable to to create a product for if it's too expensive for us to accommodate some super longtail risk in in some like

specific region or in some new sector um relative to our operating expenses then we're just not going to be able to offer insurance and I think what like we've seen with with defi largely today in its ability to accommodate longtail markets the same effect is carried over into insurance as well when you decentralize the core functions when you when you kind of codify those administrative expenses to make them uh like smart contracts you significantly lower the like theoretical overhead of offering insurance and and in doing so like like greatly expand the applicable markets that you can offer Insurance to so you you essentially open the door to a bunch of new longtail markets um that a large inefficient kind of existing or or

incumbent insurance company would have a more difficult time offering coverage for yep yeah I mean there's uh there there was a big problem there with um the insurance companies and and the bull went when everything was good but then the hacks kind of happened and people stopped wanting to provide cover and um yeah just overall was kind of a Slowdown there in the insurance realm do you think that's going to pick back up like significantly in the future and clearly smart contract risks are you know a problem amongst others you know is the future bright for crypto Insurance protocols in defi insurance or is it kind of like a tough tough uh industry to crack as far as obtaining

Underwriters yeah I I I think I think it's bright I think insurance is uh I think the second largest kind of pool of capital behind Pension funds in in the US at least and it is a like a very sizable industry that that does guide a lot of our economic market so I I think as crypto matures as a like a financial sector so will Insurance just naturally um I think there's been a lot of information asymmetry that's led to to sort of the The Limited growth of insurance in the space I think there's like generally just a a like a lack of risk aversion in the space today I mean we're all kind of degens just throwing money around in random contracts so insurance tends to not be the first

thing on on a lot of kind of existing uh base of crypto's minds I think as institutions get more involved I think as as crypto matures as an industry we'll see a like a a more well-rounded perspective in in kind of considering insurance or or hedging in in the kind of daytoday of of the like Financial risks that people are taking so when it comes to longtail assets and ensuring like like what happens with those what what is the general approach when you when you take a look at a longer taale asset how do you how do you consider it how do you value it how do you ensure it yeah it depends on the on the risk quite heavily I think this is where AI

like plays a a big role because AI is is the most deflationary technology we we've like ever had access to it allows us to make sense of massive amounts of information in a highly sophisticated and a highly efficient way and so in these longtail markets where we had kind of spse data we had um a kind of like lack of of track record or history to go off of we can make sense of of like increasingly sparse data sets um in increasingly kind of like accurate or intelligent ways and so I think it's a matter of of leveraging the these kind of new tools that we have access to now in kind of advanced AI models Advanced Quant models um to to uh develop a a

kind of more comprehensive perspective around longtail markets um so I think that's a lot of what goes into it got it it it's so cool to like start to see like how these these like two technologies that like bring Capital efficiency to the market start to intertwine and interplay together and and it's still yeah like it's it's still like just kind of like emerging um so I'm curious like what like what's your vision at upshot like as it it sounds like to to accurately value some of the some of the assets that don't have a lot of data

or very clear valuation models we can use AI to to start to draw some parallels and conclusions about the value of these tokens H how are you incorporating that into the model that you guys have at upshot yeah so that that's a lot of like what our our kind of core models do is is we start from this kind of valuation problem is we have a a kind of incomplete perspective on what the value of some longtale asset is let's fill in the gaps using kind of the historical data sets we have access to sentiment the um like semi-reliable orderbook data we have access to a bunch of this different data that we're we're kind of

developing capturing and enhancing frankly through our models um and then we we kind of develop this value for for these different assets and then with a a more reliable notion of value in hand you could start to treat them at least like a bit more like a a more fungible or liquid asset start to build kind of sophisticated Market making strategies trading strategies lending strategies I think in general where defi has really like found a a level of product Market fit is in the longtail it allows us to remove a lot of the overhead that goes into kind of active management agement of capital I think a good example is just amm in general previously where

Market Mak requires some uh knowledge of Market making the capital as well active management of Market making strategies infrastructure Etc in order to provide liquidity to any asset which all these things kind of add to the overhead of that operation amms via Smart contracts allow Market making to be as inexpensive as as sending gas to deposit liquidity into a smart contract and so now th those expensive operations that limited the number of assets we could provide active Market making strategies uh for expands like considerably because now it's so inexpensive to provide liquidity to an asset as we really start to get into the

longtail curves and and pricing logic that can be codified on chain is just going to be too minimalistic to support a lot of these assets and so we need a more robust um Tech technology as AI um to to sort of realize the the same kind of value that amm have brought to a lot of kind of longtail defi assets um xyk doesn't apply to nfts as as easily as it does to a fungible asset and so having an ml powered pricing rule that can be uh interacted with on chain allows people to uh uh create circumstance or create situations similar to that of amms oneclick Market making provide liquidity super inexpensive ways to a

much larger set of assets and do this across the financial stack spot Market making I think is just one example yeah and isn't uh isn't the the beauty of using like an AI tool is that a lot of the the value that comes with something like an nft or even any of these arbitrary assets uh on chain can be very subjective based on someone's taste um where the AI and ml doesn't really have a a subjective preference for uh colors or a different schematic that like hits your brain with with some dopamine when you look at it so that kind of AI modeling it removes the human

error aspect and also creates a more transparent and like Fair playing Ground yeah yeah and I I I think at the end of the day like like humans are all are are people are always going to be the ones interacting with markets that subjective taste is always going to be revealed through market prices different types of Market interactions and what's nice about AI is is it can look to this kind of great Arbiter of Truth we have which is the open market and and uh uh kind of understand these subjective tastes around things that that may not be kind of immediate immediately clear by just kind of analyzing Market data so so yeah I I agree and I think like still using the underlying Market as as this

really robust kind of arbiter of Truth for developing a more kind of nuanced perspective of of kind of new types of assets is still a core piece of a lot of this I wonder if if AI can detect some of the patterns that like really like provide lots of dopamine kind of a double I I think so I I mean any any like data that we use to like understand like what what drives dopamine what what doesn't drive dopamine AI is able to like get access to that that kind of data and make sense of it in just like a a more sophisticated and efficient way so yeah I definitely think

so cool so Switching gears a little bit I'm curious uh like is the appetite for some of these Market making strategies and these pricing models is that in demand mostly on the the institutional side or is that more of a retail demand that you've seen in the market um I think to date nfts have just like biased slightly to retail in general so I think that's where we've seen a lot of longtail appetite um but I I I think there is like an opportunity for like considerable returns as as people deploy in in longtail uh assets so I I I think we'll start to see more institutions enter it I think we've seen

this with defi largely over the past three years is is an appetite to get access to kind of long longtail uh type return by interacting with long T markets if if the infrastructure is there and so I think we'll start to see uh like more institutions uh get into these these especially longtail markets as the infrastructure matures a little bit more very cool awesome is there is there anything like uh that you find particularly exciting or or really compelling about the industry in Defi and nfts aside from upshot yeah I I mean this is this might

be a little biased but I I I do think that like we're starting to see more of this across the space of of bridging Ai and crypto in a lot of ways I think obviously like AI has been at the Forefront of of like most people's minds over the past year plus um and it's been getting a lot of attention but I I do think there's like a a part of AI that is is kind of immediately applicable to crypto today in these kind of narrow AI sector or like um I guess like more structured models AI like while generative AI like these kind of Transformer based models are are like really fun to play with and and really interesting and they're progressing at a really fast pace they're not quite

production ready yet for a lot of different use cases I think we're still trying to figure out like where generative I slot generative AI slots in um how it can be used how to make it sort of as robust as it needs to be but more structured and and narrow models I guess are used in in most parts of the financial system today and they're much more tractable um kind of Technologies to build in more trustless and verifiable form factors and so I I think we're starting to turn a corner in bridging AI to defi in a really meaningful way and starting to see these kind of focused experiments around how AI can increase yields across defi enable fundamentally new Financial Primitives and over the next year we'll

start to see a lot of this stuff come to Market in really meaningful ways and austr a kind of like D5 3.0 of of AI kind of marrying defi um and and a like a new design space entirely emerging as a result of that yeah very cool yeah Rob that's the uh very similar to what uh Mosaic is doing they've got AI optimized bolts where the AI kind of chooses the strategies based on what it sees in the and what kind of data it's processing yeah and there like those targeted approaches are only starting to like emerge like the the large language models the the generative AIS like

obviously the chat gpts like those do so well with language and and uh like literary Styles but you ask it a math question and it struggles like you gota you gotta tell it to like double check its work and stuff so it it makes exactly and that there like like you're saying Nick like there aren't quite targeted models that are structured for data sets for numbers um so hopefully like we start to see either like a a solution that is able to handle both or there's like more optionality that comes out where you can you can choose a a generative AI model that is targeted for

numbers and data sets yeah for sure for sure I think we'll start to see a lot of this I I think like this this like new field of zkl has started to emerge as well um and the the types of models that are es especially kind of capable at dealing with financial data are are kind of tree based models um instead of uh uh kind of like these generative large language models and the the benefit of that as well is is they're considerably smaller as well and so they're much easier to kind of build inside of a a ZK circuit or a ZK proof system and I think that is a big piece of bridging AI to crypto AI remains this black box that is this kind of trusted party in in

anything that it interacts with and in order for it to like be a viable kind of piece of a decentralized ecosystem it needs to be built in like a more trustless and verifiable way we need to verify that a price or a strategy that is the result of of some AI model then uh it is actually the result of that it is kind of verifiable um to a larger degree than than what we have access today yeah I mean that's a wild thought that I've never really even thought of right it's just crazy that's the AI models that we have say the llms and the others are are all just operated by some entity um Sam alman's World coin eye scanning orb is gonna steal your girlfriend in the

metaverse yeah like that's just we just can't have that you know yeah yeah I think if we if we really want to like build a decentralized financial system and just like a a decentralized world that is as capable as as the traditional World in these Clos systems we need to bring the kind of capabilities of of closed AI to to crypto in a like a a verifiable and trustless way I think that that's incredibly important for us like realizing this future yeah yeah totally how about everybody that left crypto for AI in the bear now we're about to pump their faces off and they're going to come back and buy the top huh yeah yeah exactly they're going to come back soon for

sure oh man there there's a few other projects that have just like started to like break into that like intersection of AI and crypto like uh there's there's one uh there's one team that's running a Linux VM uh and they built chain GPT so it's it's the the verifiable uh link between input and output so you can trace okay this input generates this output because that proof is generated on chain um there's another project called Uma that's doing like uh like fact checking uh so from a large language model you can basically like call a dispute um and then that goes

into this dispute verification mechanism and it'll tell you like based it'll kind of ask like a a bunch of human verifiers to see like was this thing real is this thing misinformation give you a give you a rundown there then the last one that I want to mention is a a I swap I this is kind of like lowkey I I heard about it at a hackathon and it's a large language model plugged into Unis swap so instead of selecting uh like token a token B you type in what you want to do swap from this token to this token or Bridge from this chain to that chain um and it it does it for you like the large language model interprets the the language of the transaction and then executes that

transaction on chain wow yeah that's really cool I think I think AI will will kind of uh be used in these like new intent based paradigms really heavily I I think like intents and AI are kind of a perfect marriage I think that's sort of the like the first meaningful use case of generative AI kind of entering the the crypto space is is via intent yeah that's an interesting one well I think we're coming up here on on Time Nick where where can our community get involved D and uh joined the upshot Revolution yeah I think uh upshot doxyz the the website is is the best place to start um jump into our Discord there's a

lot of stuff happening around um kind of bridging Ai and crypto and how to think about kind of uh like what new defi Primitives what new defi Innovations become unlocked when when AI becomes a like a more usable primitive in that space I check that out got that XYZ thanks thanks a lot Nick for coming on it's been a pleasure yeah thanks for having me guys cheers thanks Nick