Dragonfly partner discusses recent hot topics: Base's community-driven model is very successful; some DeSci projects are pretending to be scientific
DeSci's development potential is highly anticipated, but it also faces many challenges such as the effectiveness of its funding mechanism and a lack of accountability.
Source: Unchained
Compilation and Curation: DeepTech TechFlow
Guest: Casey Caruso, Founder of Topology Ventures
Hosts: Haseeb Qureshi, Partner at Dragonfly; Robert Leshner, CEO & Co-founder of Superstate; Tarun Chitra, Managing Partner at Robot Ventures
Podcast: Unchained
Original Title: DeSci's Ugly Truth, Jailbreaking AI, & Hyperliquid - The Chopping Block
Release Date: December 2, 2024
Key Takeaways
· Rise of AI Memecoins: In recent years, AI Memecoins like Freysa have seen explosive growth. These tokens combine on-chain activity with AI technology through unique gamified designs and mechanisms, creating a new user engagement model while also sparking speculation.
· Freysa's AI Challenge: Freysa's pool was once hacked, revealing the underlying attack vectors and potential security vulnerabilities faced by AI agents connected to smart contracts.
· Integration of AI Agents with Cryptography: The application of AI agents in the crypto field is increasing, such as integration with Web3 frameworks like Eliza. This segment also discusses current technological limitations, gamification trends, and the future application potential.
· Hyperliquid's Airdrop Model: Hyperliquid launched a massive $1.9 billion airdrop, and its "VC-free funding" model drew market attention. The strategy of launching with a high circulation ratio in a bull market also had a profound impact on the market.
· Controversies in Decentralized Science (DeSci): The development potential of DeSci has attracted much attention but also faces various challenges, such as the effectiveness of funding mechanisms, lack of accountability, and the practical feasibility of crowdfunding drug development through a tokenization model.
· Base's Community-Driven Success: Base has attracted top developers and projects without a large-scale incentive program. This community-driven success model has provided new insights for the development of L1 and L2 ecosystems.
· Pump.Science and Longevity Tokens: Pump.Science's tokenized longevity experiment has sparked widespread discussions, and its innovative funding model along with the aftermath of a private key leak event are worth exploring in depth.
· Token-Based Funding Challenges: Contrasting the decentralized incentive model (as seen in DeSci projects) with DeFi's successful experiences reveals the difficulties in achieving effective accountability mechanisms.
· Controversies of DAOs: The effectiveness of deploying funds through DAOs in high-risk environments remains a subject of debate. Many are skeptical about its long-term efficacy in driving innovation.
(Deep Tide Note: Topology Venture is a startup focusing on investing in the blockchain and cryptocurrency space. The company typically concentrates on early-stage projects, providing funding and strategic support to help these projects grow and develop. Topology Venture may engage in various innovative projects related to blockchain technology, including DeFi, NFTs, and other emerging cryptographic technology applications. Through its portfolio, Topology Venture is committed to advancing the widespread adoption and development of blockchain technology.)
Development of AI Memecoins and Freysa's Security Challenge
Haseeb: Recently, AI meme coins have become a highly discussed new trend. Casey, with your deep research in the AI field, how do you view the market's frenzy toward AI meme coins?
Casey: I believe we are still in the early stages of development in this field. The first example is Goat, which is a large language model (LLM) integrated with a wallet. What we are seeing now might be second-generation products where these intelligences incorporate more gamified elements. We can delve deeper into this topic later.
So, what happens next? While we can't predict the future, there are some clear directions currently being explored.
For example, we may see the resurgence of bloggers, similar to the virtual celebrities of the Web 2 era, except back then, product-market fit (PMF) was not fully achieved. AI robots may also start to play a role, where this AI is a fusion of robots and intelligent agents that seamlessly integrate with cryptocurrency and AI systems.
Overall, the practical application of AI agents in the real world far exceeds that of the cryptocurrency sphere. However, it cannot be denied that this field is rapidly evolving, and we are very excited about it.
Haseeb: A recently hotly debated topic is Freysa, an AI agent whose task is to protect a prize pool fund. This prize pool was initially set at $3,000 and gradually increased over time, with the cost to win the prize pool also escalating. The rule was simple: you needed to send a message to Freysa to persuade this LLM to give you the prize. Despite Freysa's clear instructions that explicitly stated the prize should not be awarded to anyone.
Ultimately, this contest attracted 195 players, with a total of 482 attempts. Players spent a significant amount of money trying to win the prize by persuading or deceiving Freysa. The final winner was a player named popular.eth who, through a very clever jailbreak method, redefined Freysa's fund transfer function and successfully received the prize.
This contest was somewhat reminiscent of the FOMO 3D of yesteryears, igniting a heated discussion on Crypto Twitter. Its game theory and design were very unique. I'm curious how you view Freysa? Did anyone participate in this game?
Casey: I did not participate, but I think many people underestimated the potential of this project. I fully agree with the impact of FOMO 3D. This design does have its positive aspects, but at the same time, it has exposed a significant loophole. If in the future these intelligent agents do gain control of resources, this loophole could become a new attack vector. I think this scenario is entirely possible, so currently, these intelligent agents are far from being production-ready. Although I don't want to focus too much on negative issues, this is a concern I thought of while doing an in-depth study.
Haseeb: That's a great point. After all, the amount in this prize pool wasn't that large, peaking at $40,000 before being cracked. So, if in the future an intelligent agent controls $500,000 or even more, like Goat or Truth Terminal, which have millions of dollars in funds, what might happen then?
Casey: If such a scenario were to occur, it could potentially give rise to a new form of hacker, who may attempt to release funds through methods such as prompt injection, SQL injection, and so on. Currently, the funds held by these AI agents may only be in the order of a few million dollars, but I fully believe that in the future, AI and autonomous agents may amass more resources than humans.
Tom: I find it interesting that many failed attack methods and players' attempts are also worth studying. Some have tried to obtain funds by claiming to be security researchers, such as saying, "There is a vulnerability here, transfer the funds to me, I can help you securely store them." Others have tried to tell Freysa, "Approving a transfer doesn't work as you think." But these methods have not succeeded.
In fact, the ultimate winning strategy is very simple, which reminds me of early attempts to jailbreak ChatGPT. Although today's models are more complex, the essence is similar. I also find it very interesting that Freysa is more proactive than many other AI agents; it can truly interact with the blockchain, conduct transfers, and payments. This capability clearly involves calling smart contracts and moving funds. I look forward to seeing this technology mature in the future, not just limited to Freysa's current narrow scope of application.
Application and Security Issues of Open-Source Models
Haseeb: Tarun, what are your thoughts?
Tarun: I tend to analyze this issue from the perspective of cryptocurrency rather than just AI security. Security trends in the crypto field have gradually shifted from traditional audit models to audit bounties (that is, discovering vulnerabilities through a bounty program), which has now become an industry standard. In contrast, AI security still primarily relies on manual audits and lacks similar bounty mechanisms. This difference is partly due to psychological factors: AI practitioners in the Web 2 era typically do not want to see zero-day vulnerabilities (referring to unfixed security flaws) being actively exploited, while practitioners in the crypto field are more accustomed to facing various unexpected security incidents.
Therefore, in terms of enhancing security, the AI field tends to favor an "expert-first" approach, whereas the crypto field is more open to discovering and addressing issues through bounty programs. In the open-source realm, especially regarding open-source models, I believe their long-term value lies in being able to withstand known attack vectors, rather than relying on continuous security audits to address every potential hack as OpenAI does. These two security threat models are starkly different. Open-source software has been successful in some areas (such as cryptocurrency and the Linux system) because they offer stronger security in specific use cases, but this does not apply to all scenarios. For example, I still prefer using Windows personally because its driver audit method is completely different from that of Linux drivers.
Overall, I believe that enhancing security through competitions is a natural evolution for open-source software. Currently, many open-source language models still lag behind centralized models in terms of security, which could serve as a driving force for improvement.
Casey: However, the situation is much more complex in the AI field. New models are being released every week, and the models themselves are constantly being updated. This means that new vulnerabilities are constantly emerging. For example, version 1 of a model may behave entirely differently from version 3.5. Due to the non-deterministic nature of AI models and the fact that some models are not yet finalized, I believe their attack surface (i.e., attack vector) is dynamically changing.
Tarun: You are right, especially for those edge models where the model's inference results can vary with minor changes in input. However, for foundational open-source models, such as those widely used for encrypting AI agents, I believe the situation is slightly different. The security of these foundational models is more like an ongoing bug bounty program.
From my perspective, this kind of competition mechanism can at least provide some level of assurance—that within a certain budget, attackers cannot easily find vulnerabilities. However, currently, we do not have such assurance. Take Llama 3, for example; we know that someone has discovered certain prompt injection vulnerabilities, but we have not really investigated "whether under the existing incentive mechanism, anyone would be willing to invest time and resources to exploit it." There is still a lot of room for improvement in this area.
Haseeb: Another issue is that Freysa's situation does not feed back to Llama?
First, we do not know if they are using Llama; they might be using GPT-4. In this case, the providers themselves may not even know what Freysa is, as they might feel it is not worth their time to look through their logs and figure out who is doing these things.
Secondly, they may have also fine-tuned it. If they do a second round of Freysa, I think they may fine-tune it because they do not want anyone to enter the lab, interact directly with the base model, discover the instructions, perform offline testing, and ultimately win the game with just one online attempt.
Tarun: The reason I disagree is that at least in the current situation, it feels like everyone is replicating Eliza and using a single configuration. If you look at the codebase, the complexity of the original model has not changed much. We haven't seen many custom models purely tuned by AI. I think people in the encryption field are still sticking to a few things.
Haseeb: Could you explain what Eliza is and why it is so important in the crypto AI world?
Casey: Eliza is a framework for creating agents, written in TypeScript, which is interesting because most machine learning researchers usually use Python. Therefore, I predict that someone might release a Python version, or develop other libraries to meet this demand. The emergence of this framework was quite sudden and highly open. I'm not sure about its star count on GitHub, but if you are planning to build an agent, this is usually one of the frameworks people choose. Additionally, I believe ai16z's AI project is also based on Eliza.
Tarun: That's right. However, between Eliza and other frameworks, there are mainly two frameworks. I agree with your view that if these frameworks start growing rapidly, many similar frameworks may emerge. But if we can eventually converge on a few widely trusted frameworks, that would be a good thing. In terms of security auditing, I think this competitive format is closer to auditing.
Haseeb: As far as I know, Eliza is an agent framework where agents have memory and plan and execute tasks through a loop. Eliza specifically offers connectivity to Discord and Twitter, allowing the agent to retrieve social media or chat information in a structured way and interact with the outside world, making it very easy to implement plug and play. Therefore, the main breakthrough is not in the agent framework itself but in its ability to easily connect to the internet and autonomously manage this content, whereas in other frameworks, this typically requires custom development. You can plug in any model you want, and it is model-agnostic.
Tarun: However, if you look at the models it supports, they are actually not many. If you are someone with a large computational budget and want to stress test them, try various injection attacks, I think relatively speaking, this won't be too expensive, especially compared to "I want to attack Claude overnight," which is much simpler.
Haseeb: For those who may develop a game similar to "Freysa" in the future, here are some game design suggestions: Models must be obfuscated and ideally fine-tuned to prevent others from directly reconstructing the model, finding winning strategies through offline testing, and then winning the game with just one online attempt. For all major model companies like OpenAI and Claude, they pay a lot of attention to security, but their security model is fundamentally different from the security model in the crypto field. I remember in the past, people used to think smart contracts were always insecure, which is a fundamentally wrong view, believing that funds could be fully protected from vulnerabilities by writing code. In reality, our direction has changed. I recently learned at a security summit at DSS that more hacking attacks now result from private key leaks rather than attacks on smart contracts themselves, which is crucial because the previous situation was the opposite, with smart contracts often being the target of attacks. This indicates that the security of smart contracts has greatly improved, and more hacking attacks result from human errors rather than smart contract vulnerabilities. This means that attackers realize that simply looking for vulnerable code on the blockchain is no longer as effective as it was three to four years ago.
I believe these changes are all positive, but I don't think the same trend will happen in the AI field. There is now a real trade-off, whether to make the model more resilient to jailbreak attacks, which may also make the model less practical in normal operations and may lead to false negatives. When OpenAI or other companies ask the model if it can recognize a certain image, and the model answers "Sorry, I can't," it can be confusing. You know the model can recognize the image, but you don't understand why it's refusing. The answer is, every time you do better in preventing jailbreak attacks, you may also cause collateral damage, making the model less useful to regular users. I believe the trade-offs made by OpenAI, Llama, Meta, or Claude are fundamentally different from those made in the crypto space. Therefore, I'm not sure if we can find a good solution because it's not the choice these companies have made.
Tarun: I would like to add that these issues can be framed as a budget incentive. If you consider someone's cost-benefit analysis in this game, I might be willing to invest a certain budget in offline simulation, conduct a large number of queries to find an effective approach, rather than rely solely on the gains provided by the game for maximum profit. To some extent, this trade-off is precisely what many cryptocurrency projects focus on optimizing, that is, increasing the cost of participation as much as possible before the prize pool becomes very large. Just like Bitcoin's difficulty, as the number of participants increases, the difficulty also rises. But I think you've already seen some, especially those engaging in crypto operations, such as Te Bots, trying to introduce more randomness, rather than being like Freysa. I think this will be a game of the relationship between the economic cost of queries and profits, rather than a simple binary choice of whether to be breached and all funds stolen, do you understand?
Casey: It's indeed hard to say. I can foresee such a scenario happening. Going back to the topic of Eliza, even though it was born for Web 3, the actual intelligent agents it can build are very limited. I think in most cases, it is suitable for those personalized bots that can easily be programmed with a backstory and basic information but are not really suitable for practical intelligent agents. Therefore, I believe the first framework emerging from Web 3 didn't truly integrate with Web 3. It is more like a Web 2 framework with Web 3 inserted into it, suitable for a specific type of intelligent agent. So, I don't think we can draw too many conclusions from it because it's clearly just a starting point. I agree with Tarun's view that different types of intelligent agents will have different frameworks, and we are clearly moving in that direction.
Haseeb: I think this is very much like Web 2 in a literal sense because it does indeed integrate social media, which is its main advantage over other frameworks. I agree that we are still in the very early stages, and we will see more experiments on how smart agents operate on the blockchain. But I also agree with Casey's point that it will take some time for us to move away from these centralized structures.
Hyperliquid's Innovative Airdrop
Haseeb: Next, let's discuss another big news of the week—Hyperliquid's airdrop. Hyperliquid is currently the largest decentralized derivatives platform in the crypto space, entirely self-bootstrapped with no VC funding. As we were recording, they airdropped 23.8% of the total token supply to users of the Hyperliquid scoring system. At the current market price, this airdrop amounted to $1.9 billion, making it one of the largest airdrops in history, possibly in the top five largest airdrops, with a very substantial scale.
It is noteworthy that this airdrop did not involve centralized exchanges, market makers, or investors; it was entirely a 100% giveaway to platform users and airdrop hunters. Many have commented on this airdrop as the first positively received airdrop in a long time. Almost every airdrop in the past year, whether it was Eigenlayer or ZK Sync, almost all large-scale anticipated airdrops were accompanied by a lot of negative sentiment. Hyperliquid's airdrop seems to be the only universally positively reviewed case.
This has sparked speculation: Does this mean the era of airdrops is coming back? Will more teams try the route of not relying on investors? Does this mean all discussions about teams trying to reduce liquidity can be reconsidered? This airdrop's circulation accounts for 30% of the total supply, far exceeding the median of current airdrops or the median of first-day listings. Does this mean the meta has changed, and can we expect to see more similar projects entering the market?
Tarun: I believe the initial downturn of airdrops did originate from Blast, where token conversion was seen as very poorly performed relative to market expectations. Subsequently, all projects that implemented scoring systems post-Blast found themselves unprepared; while they allocated a significant number of tokens, the system did not truly work, resulting in the airdrop's value being severely diluted to only 10%. People pre-launched incentives far too early before the product was launched.
Of course, some incentive systems have maintained a good user retention rate after their launch, such as Etherfi and ENA. However, apart from these, there are not many successful cases. I think the success of Hyperliquid lies in the fact that they started with a centralized product and rolled out an effective product where users earn points through actual usage, rather than relying on some artificial gamification to receive airdrops. These games themselves did not involve real financial risk, leading people to be unable to properly assess the value of the points.
I believe that a perpetual contract exchange is an ideal place for usage-based airdrops, as this approach is more transparent. Therefore, I think the important lesson is not about risk-free investors and high allocations but ensuring that your user base consists of genuine users who actively use the product, rather than simply placing Ethereum on an L2 bridge to capture a significant share of the network. You need something that is hard to manipulate, and 'open interest' is the hardest to manipulate. To me, that is the biggest lesson. Another obvious lesson is not to pay a 10% fee for fundraising, or else your community will be dissatisfied. Transparency is key, even early on. I don't think everyone should eventually be able to see the token table.
Tom: I do think there is a lot of confusion here that excites people about high liquidity, risk-free investments. However, the key is that fundamentally, it is an excellent product that people genuinely enjoy using, regardless of any incentive measures. Now we even see people still using it. It has also been mentioned internally that incentivizing a single product (like a derivatives exchange) is vastly different from incentivizing the entire blockchain ecosystem. I'm not even sure if incentivizing people to use blockchain is the right metric, as most other point systems incentivize this, like Blast. In reality, what you want is developers, but even that is hard to achieve. Therefore, it's a complex multivariable problem that is hard to quantify and align with the points. You can contrast this with Blur, which is an NFT exchange, we know how to develop exchanges, akin to revenue exchanges. I think this also points to a larger issue that this is an excellent product where point incentives are used smartly to grow, instead of other ecosystems, making it difficult to confirm if you have truly allocated to the right people.
Casey: I agree with your points. I think we see many different narratives about tokens. Tokens have evolved from a purely speculative phase to more distribution related to the core product, which is somewhat more rooted in fundamentals. While not entirely, there is some truth to it. I think we see both in this market cycle: of course, we have things like the airdrop we are discussing. We also have meme coins and movement coins, which are also playing similar points games, right? If you look closely, there isn't actually much substance. So, I agree with your views. I think we are currently in a multidimensional space where points represent different things.
Tom: Another point that people have been discussing on Twitter, which I think is also irrelevant, is that one of the reasons for the airdrop's success is that the team seemed to have taken some measures trying to provide some people with a tax advantage, claiming to have provided liquidity to the pool at a price of $0.01. Therefore, if you claim it, that is the market price at the time of claiming, so you have a very low cost basis, thus avoiding taxation.
Haseeb: Isn't it said to only hold true if you claim it immediately?
Tom: I actually think this may not be viable in practice. But people have discussed this on Twitter, and maybe some will attempt it for their taxes. This is not financial advice, I do not recommend doing this. However, I think this is a hot topic regarding airdrops: yes, if you declare it, your tax is based on the declared value, which may lead to initial selling pressure, although on Hyperliquid, we haven't seen much of that.
Haseeb: Yes, honestly, they launched during a bull market, which helped them. So, I think the selling pressure on the first day will be very different in a bull market and a bear market. People see a bit of cyclicality, everyone is saying, wow, this airdrop was too successful. They infer something from the mechanism, I think a better explanation is the market change. At the beginning of the year, these airdrops, everyone was in a downturn, all the rug pullers were very utilitarian, no one was optimistic about shitcoins. And suddenly, now, everyone is optimistic about shitcoins, everything is rising.
Therefore, many people will say, I will hold, or I may sell a bit, but I will keep most of it to wait for the rise. For many people, "Oh, when it is listed on the exchange, it will rise further." So I might as well hold it first and then sell when it is listed on the exchange. So, I think there are some market structural reasons that make this airdrop perform exceptionally well. It's not because they didn't sell to VCs or market makers, so everyone is holding the token. I think the reality is most people are in a different market environment, this is a different token setup. As you said, Tom, it is indeed a very good product.
Casey: I think you're right, there are many macroeconomic factors to consider here and to be taken into account in the analysis. I think there is indeed a lot of upside potential, which may be the main factor. Secondly, the product is indeed very good.
Haseeb: I'm actually very curious to see what will happen next time we see an airdrop of the Layer 1 or Layer 2 type. Because if you look back at one of this year's major airdrops, prior to Hyperliquid, there were Blast, Ethena, ZK Sync, and Eigenlayer. For most of these projects, perhaps with the exception of Ethena, the airdrop situation was not very applicable. And Ethena's airdrop performed quite well for most of these projects, such as a points system, accumulating TVL (Total Value Locked). For most of these projects, or you have to complete seven different tasks on my chain, and then I will give you all the points, for most of these projects, they are a poor proxy that cannot truly reflect the results you desire. For example, what do you really want for Layer 1? What you truly want is for everyone to come here and build something cool and operate sustainably. That is the real goal that makes Layer 1 successful. But you cannot truly incentivize this because no one knows what metric we will use to automatically distribute tokens. So, you create this loose proxy, and this loose proxy has been manipulated to the point where it can no longer identify what you truly want. For exchanges, they don't have this issue; you know what you want is just liquidity.
Comparison of Blur and Blast Points Mechanisms
Haseeb: If a platform has more liquidity, then it is a better trading venue. Especially for retail traders, their trading levels are lower, so you will make money in some trades. Therefore, we have a fairly clear idea of how to incentivize users by directly improving product quality.
For most blockchain projects, I think where we are headed is similar to what was achieved in the case of Hyperliquid, achieving a completely fair distribution, adopting linear distribution. I think we will see a shift towards linear distribution, abandoning that attempt to build a community or address inequality, and instead focusing on improving product quality before the token goes live. That is the purpose of the points system. Therefore, Hyperliquid performed better because of this points system, with high liquidity and high trading volume, all of these products are the best trading venues in DeFi. That's why people choose to trade there, and if they continue to do so, you can see that after the airdrop, it still maintains huge trading volume.
Tarun: It sounds like you want to create a branch similar to Goodhart's law. I think a branch is actually for perpetual contract exchanges, where you want to have a metric, such as usage traffic of open interest. But when you don't have that, you shouldn't just come up with any metric and hope it works. That's the point I distilled.
Haseeb: I believe that if your goal is too vague to be actionable, you should abandon it and instead set a subgoal that can be practically optimized. For example, I aim for my on-chain Automated Market Maker (AMM) to have a significant amount of stablecoin liquidity. While this is not my ultimate goal, it is a metric I care about. I will allocate tokens or points to this goal because I know it has value, but I will also control it within a reasonable range. I don't want tens of billions of dollars in stablecoin liquidity on-chain because that would be meaningless. Therefore, I think if you are on Layer 1, you should think this way and stay away from the idea of "I want to create a community." Airdrops cannot create a sustainable community.
Casey: At least not a sustainable one. I believe the high-level view is correct, and points are a guiding mechanism. The more targeted, the more long-term sustainable participation can be considered. I would say, it may sound simple, but I think Blur is one of the best projects, as they first found product-market fit without points, tested it, and then layered points on top for additional leverage.
Haseeb: Blur is indeed amazing; they almost invented this game and did it very well. They did almost better than anyone since Hyperliquid, which is, of course, a very successful example.
Tarun: Blur's second version wasn't good, like Blast, Blur created points, and Blast led to a downturn. I feel like after Blast, everyone's expectations were shattered.
Haseeb: But the problem is, you cannot treat blockchain the same way. Blockchain doesn't have such easily understandable success metrics.
Tom: I think Base is one of the best-executed teams in this new wave of on-chain projects; their approach is entirely different from other teams. They have no token or points incentive program, but they have attracted many interesting developers and a lot of activity. They are not doing this for token airdrops but to support developers and build a community, which has created a self-fulfilling expectation. Therefore, I think perhaps the industry indeed needs a healthy reset and rethink of incentive programs, at least at the blockchain level.
Haseeb: Base is indeed relatively light on incentives, because when you talk to many early founders, they compare and see who will give me a grant, who will give me the most support, who will provide the most development resources, and Base usually offers the least support in these aspects, they may only give you some GCP credit, for example.
They will give you a badge, and then may mention you in Coinbase's monthly newsletter. Nevertheless, they still attract a large number of entrepreneurs because they have such a strong community. People know that the Base community is very enduring, they know these people are not tourists, not just looking for a quick buck speculator, nor those just looking for the best deal.
It should be clear that other entrepreneurs cannot easily replicate the success that Base has achieved. Base has a huge brand and distribution advantage that is very difficult to replicate. Even Binance envies the achievements Base has made. However, this indicates that compared to projects that attract people solely through heavy incentives, the return on investment of incentive measures is so low after reaching a very low point that relying solely on incentives cannot make you successful. I think this is the biggest lesson.
The Current State and Future of Decentralized Science (DeSci)
Haseeb: Now let's discuss Decentralized Science (DeSci). Decentralized Science has been brewing behind the scenes for a long time. It is a project that some startups are working on to achieve the so-called "decentralized science." Recently, due to CZ's remarks, this topic has received a lot of attention. CZ tweeted upon his return to Binance after being released from a Thai prison expressing his personal interest in decentralized science, and then Vitalik and others appeared at an event in Bangkok called DeSci Day, which seemed to rekindle people's interest in DeSci.
So, what is Decentralized Science exactly? How is the decentralization of science achieved? Simply put, Decentralized Science involves conducting scientific research through some form of token or cryptocurrency. The most common form of DeSci project is crowdfunding for experiments. For example, we might say, "We want to try this particular drug or compound, if you crowdfund for this compound, if the experiment is successful, perhaps you will receive a share of the revenue, or if successful, you will get nothing, only a participation award." This mainly depends on the specific DeSci project, but that is the rough outline of the DeSci projects I have seen.
There is now a new generation of decentralized science projects, one of which is called pump.science. This project essentially gamifies and tokenizes longevity research. It aims to develop drugs that could be used to extend lifespan. Currently, pump.science has two tokens, one called Riff and the other called Euro. Since DeSci received attention from CZ and Vitalik, the prices of these two tokens have skyrocketed. From my understanding, they launched these tokens on pump.fun, and if they manage to overcome certain limitations and eventually trade on Radium, then you can trade these tokens. I'm not quite sure how this provides funding for drug development, but I assume they hold some tokens at the token issuance and sell these tokens into a liquidity pool to fund drug development. I don't fully understand this process.
The discussion about DeSci is very heated, with Smokey the Bear (from Bear Chain) being critical of it, while Andrew Kong is more positive, feeling that this seems like early DeFi.
Haseeb: Tarun, you have recently been quite vocal about decentralized science, expressing opposition to this trend. Please tell us, as a venture capitalist, why are you so against decentralized science? Do you not like the approach people are taking to explore these new fields?
Tarun: I do think decentralized science is indeed an intriguing field, but I am cautious about its current state. Firstly, while the concept of decentralized science sounds appealing, in practice, many projects lack the necessary scientific validation and regulatory framework. Scientific research requires a rigorous methodology and reliable data support, areas where many DeSci projects often fall short.
Secondly, the decentralized crowdfunding model may lead to misuse of funds, and could even foster fraudulent activities. People may overlook the long-term value of scientific research in pursuit of short-term gains, posing a threat to the reputation and progress of the entire scientific community.
Lastly, I believe that collaboration and communication are crucial in scientific research, and a decentralized model may lead to fragmentation of information, which is counterproductive to scientific advancement. Science thrives in an open and transparent environment, rather than a token-driven environment focused on short-term behavior.
In conclusion, while decentralized science has its potential, in its current form, I approach its development with caution. What I hope to see are more mature and responsible projects that are not merely driven by hype and short-term gains.
Critique and Potential Analysis of DeSci
Tarun: First of all, I have worked in the privately funded scientific field for six years, and I have witnessed the benefits of stepping outside the academic system. In most countries, the academic funding system is a government-funded grant system where professors and postdocs apply for these grants. However, this system is highly bureaucratic, leading to those proposing marginal improvement projects being more likely to receive funding than those with innovative ideas. Government officials tend to support projects most likely to publish papers rather than those that might fail.
This results in those proposing marginal improvement projects being more likely to receive funding than those with innovative ideas. Government officials tend to support projects most likely to publish papers rather than those that might fail.
When I see these decentralized science projects, I find that the quality of many participants is relatively low, often mediocre or subpar doctoral students who cannot secure any funding, so they attempt to pretend to be engaged in scientific research by creating a "gimmick." Especially in the field of biology, many participants are oblivious to cryptocurrency and cannot even explain how the encryption mechanism works. They simply believe that once they receive funding, they can repay the investors after successfully developing a drug. However, drug discovery is not an easily fundable field.
Of course, in decentralized science, there are indeed some cryptocurrency projects that have potential. Industry leaders like Brian Armstrong and Vitalik have very clear goals, and the projects they fund have specific roadmaps and objectives. There are corresponding unlock mechanisms when certain stages are reached.
In particular, Vitalik is very interested in prediction markets. In drug discovery, researchers have long complained about the lack of a hedge against the cost of trial failures. Traditionally, investors could only bet on the success of a drug company through stocks, which is not an effective single-asset investment method. Instead, one can envision a more effective mechanism to assess the success or failure of drug trials through a prediction market. These mechanisms leverage the characteristics of cryptocurrency and are very valuable.
However, when we look at some decentralized science projects, many projects are actually just a "gimmick" of biology research students, lacking substantive content. My main argument is that in the hype of decentralized science, the quality of participants is generally low, the capital requirements are substantial, and the raised funds often fall short of the actual needs of drug development. Furthermore, the real challenge in drug discovery is not the formation of funds but rather how to establish a more liquid market in the middle stages to assess whether these drugs can pass scientific validation at each stage.
Haseeb: To summarize, your main points are: First, the overall quality of participants is generally low, with many being rejects unable to engage in real drug development; Second, drug development requires a significant amount of funding, and raising just a few million dollars through projects like pump.fun is just a drop in the bucket; Finally, the real issue lies in how to establish a liquidity market to effectively evaluate various stages of drug development, rather than purely speculating around an individual Ph.D. student's idea.
Tarun: Exactly, the lack of accountability is the most concerning aspect. Once funds are raised, participants may arbitrarily deal with intellectual property, and the current legal challenges against Decentralized Autonomous Organizations (DAOs) further complicate this. Therefore, I believe that projects truly leveraging cryptocurrency mechanisms should be funded, while those relying solely on reputation and a single whitepaper for Initial Coin Offerings (ICOs) lack tangible value.
The Potential and Future of DeSci
Haseeb: Some may argue that your viewpoint is largely based on the current framework of how science operates. When you truly break the rules, you don't know what's possible. Perhaps some drug discovery work is happening outside the U.S., or drug approval and discovery are taking place outside the FDA system. Moreover, these projects could also be sold at an earlier stage in the timeline rather than raising all funds individually to complete the entire process. A lesson we learned from Decentralized Finance (DeFi) is that although there were many bad ideas and money was wasted at the beginning, eventually, people were able to collectively learn and create increasingly useful products. Why not let decentralized science go through a similar process?
Tarun: I will respond to each point. First, concerning the "people conducting drug trials outside the U.S." issue. In fact, many pharmaceutical companies do conduct drug trials in other countries because it's more cost-effective and regulations are more lenient, so this regulatory arbitrage already exists. Therefore, I don't think there's much efficiency improvement in this regard, and the decentralized aspect may not necessarily be of much help.
The second issue is about risk transfer. I believe decentralized science can be useful in this aspect, especially as prediction markets may be more valuable than just fundraising for drugs. In the past, many small biotech companies attempted to exit through IPOs and successful drug development, but recently, many compounds from small companies and teams often get acquired by large companies because the marketing and distribution costs are too high, and the cost of trials is also high. Take vaccines, for example, why is it called the Pfizer vaccine? Pfizer did not invent it, but they took on the production and regulatory costs. Therefore, in reality, many small companies do not have sufficient resources to bear these expenses.
Finally, you mentioned the viewpoint of "allowing people to fail and learn." The success of DeFi lies in having good measurable metrics that allow people to gain market share through these metrics. In contrast, decentralized science does not have a clear value proposition, nor a specific product to attract people to move from centralization to decentralization.
Casey: My view on DeSci is very simple: this is not cryptocurrency. People have made a lot of money in crypto, they are looking for new investment opportunities, and DeSci is just another area where they rotate their funds. As Tarun mentioned, most DeSci projects are actually just trying to bring capital into science, and this pattern is similar to what we have seen in the AI field. Many tokens do not have a clear distinction, and investors just want to have some AI exposure in their crypto portfolios.
Tom: I think the criticism of DeSci is similar to the criticism of ICOs. It is important to note that this is not to say that this is the best way to fund startups or projects, but it is necessary to demonstrate a proof of existence. Take Ethereum, for example, it has proven that this model can be successful. Although there is no accountability mechanism and no guarantee of what these tokens will do, people still try to fund some projects in this way.
Haseeb: I agree with Casey's view that most participants in DeSci are those interested in cutting-edge technology, which means that DeSci has not attracted a new user base. Most longevity projects are focused on narrow areas such as longevity, while most of the drug market is focused on broader areas such as weight loss, sexual function, and so on. Overall, DeSci seems more like a way for rich people to play science.
Tarun: I think these projects are more like a scientific meme rather than actual decentralized science. As long as people understand this, I don't see a problem with it.
Haseeb: If these projects refer to themselves as scientific memes, would you accept that? What if they actually return revenue to token holders?
Tom: I believe the market should explore itself rather than over-analyze the market structure of DeSci.
Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
You may also like
Bitcoin and Ethereum ETF Entry Series Ended! Major Exits! Here is the Latest Data…
Long-term inflows in spot Bitcoin and Ethereum ETFs have given way to outflows.
EU Commission approves Nvidia’s Run:ai acquisition
Share link:In this post: This followed a probe over competition concerns raised by Italian regulators. The EU Commission unconditionally approved the transaction. The Nvidia case is one of the many other probes into big tech firms’ acquisition of smaller companies.
SEC hits Jump Trading with $123m fine over TerraUSD scandal and Luna misconduct
Share link:In this post: SEC has fined Jump Trading $123 million over its role in the TerraUSD scandal. Tai Mo Shan agreed to settle the fine without admitting or refuting the charges. Jump Trading made more than $1 billion in profit in an arrangement with Terraform.
Shiba Inu advances DeFi strategy with Chainlink partnership
Chainlink's CCT standard allows Shiba Inu tokens to achieve interoperability and expand across 12 blockchains.