Dragonfly partner discusses recent trends: AI agents still have room to grow toward production-grade, some DeSci projects are gimmicks pretending to be science
TechFlow Selected TechFlow Selected
Dragonfly partner discusses recent trends: AI agents still have room to grow toward production-grade, some DeSci projects are gimmicks pretending to be science
Decentralized incentive models (such as DeSci projects) contrast with the success of DeFi, revealing challenges in establishing effective accountability mechanisms.
Navigating Web3 tides with focused insightsCompiled & Translated: TechFlow
Guest: Casey Caruso, Founder of Topology Ventures
Hosts: Haseeb Qureshi, Managing Partner at Dragonfly; Robert Leshner, CEO & Co-founder of Superstate; Tarun Chitra, Managing Partner at Robot Ventures
Podcast Source: Unchained
Original Title: DeSci’s Ugly Truth, Jailbreaking AI, & Hyperliquid - The Chopping Block
Air Date: December 2, 2024
Key Takeaways
-
Rise of AI Memecoins: In recent years, AI memecoins like Freysa have experienced explosive growth. These tokens combine on-chain activity with AI through gamified designs, creating novel user engagement models—while also fueling speculative behavior.
-
Freysa's AI Challenge: Freysa's prize pool was hacked, revealing the attack vector and highlighting potential security vulnerabilities in AI agents connected to smart contracts.
-
AI Agents Meets Crypto: AI agents are increasingly being used in crypto, such as integration with Web3 frameworks (e.g., Eliza). This section explores current technical limitations, gamification trends, and future potential.
-
Hyperliquid's Airdrop Model: Hyperliquid launched a $1.9 billion airdrop—the largest in history—with its “no VC funding” model attracting market attention. Launching during a bull market with high circulating supply had profound market implications.
-
Controversy Around DeSci: While DeSci shows promise, it faces challenges including ineffective funding mechanisms, lack of accountability, and questions about the feasibility of tokenized drug discovery crowdfunding.
-
Base’s Community-Driven Success: Base attracted top developers and projects without large-scale incentive programs. Its community-led success offers new insights for building L1 and L2 ecosystems.
-
Pump.Science and Longevity Tokens: Pump.Science’s tokenized longevity experiment sparked debate, especially around its innovative funding model and aftermath of a private key leak.
-
Challenges of Token-Based Funding: Contrasted with DeFi’s successes, decentralized incentive models (e.g., DeSci) reveal difficulties in establishing effective accountability.
-
DAO Controversies: There remains skepticism over whether DAOs can effectively deploy capital in high-risk environments, with doubts about their long-term efficacy in driving innovation.
(TechFlow Note: Topology Ventures is focused on investing in blockchain and cryptocurrency startups. Typically targeting early-stage projects, the firm provides both capital and strategic support to help them grow. It may participate in various blockchain-related innovations, including DeFi, NFTs, and other emerging crypto applications. Through its portfolio, Topology Ventures aims to drive broader adoption and development of blockchain technology.)
The Rise of AI Memecoins and Freysa’s Security Challenge
Haseeb: Recently, AI meme coins have become a major trend. Casey, you’ve done deep research in AI—what do you think about the market’s current frenzy around AI meme coins?
Casey: I think we're still in the very early stages of this space. The first example was Goat, a large language model (LLM) integrated with wallets. What we’re seeing now might be second-gen versions, where these agents incorporate more gamification elements. We can dive deeper into that later.
So what comes next? While we can’t predict the future, there are already clear directions being explored.
For instance, we might see a resurgence of bloggers—similar to virtual influencers in Web 2, though they never quite achieved product-market fit (PMF). AI bots could play a role too, combining robotics and agent-like intelligence that seamlessly interfaces with crypto and AI systems.
Overall, real-world applications of AI agents are far more practical than those in crypto today. But no doubt, this field is evolving rapidly, and we’re excited by what lies ahead.
Haseeb: One topic making waves recently is Freysa—an AI agent designed to protect a prize pool. The pool started at $3,000 and grew over time, while the cost to win increased. The rule was simple: send Freysa a message convincing the LLM to release the funds—even though Freysa was explicitly instructed never to give out any money.
Eventually, 195 players made 482 attempts, spending significant money trying to persuade or trick Freysa. The winner, popular.eth, succeeded using a clever jailbreak technique—redefining Freysa’s fund transfer function—to claim the prize.
This event felt reminiscent of FOMO 3D, igniting massive discussions across Crypto Twitter. Its game theory and design were truly unique. I’m curious—how do you all view Freysa? Did anyone here participate?
Casey: I didn’t take part, but I feel many underestimate the project’s potential. I completely agree with the FOMO 3D comparison. This kind of design has positive aspects, but it also exposed a huge vulnerability. If such agents ever control real resources, this flaw could become a new attack surface. I believe this scenario is entirely possible, which means these agents aren’t production-ready yet. I don’t want to focus too much on negatives, but this is an underlying concern when studying it deeply.
Haseeb: That’s a great point. After all, the prize amount wasn’t large—only around $40,000 before being cracked. So what happens if one day an agent controls $500K—or even millions, like Goat or Truth Terminal?
Casey: If that happens, we might see an entirely new class of hackers who use prompt injection, SQL injection, or similar techniques to extract funds. Right now, these agents may only manage millions—but I fully believe AI and agents could eventually accumulate more resources than humans.
Tom: What’s interesting is that many failed attack attempts are worth analyzing. Some tried claiming to be security researchers, saying things like, “There’s a vulnerability—send me the funds so I can secure them.” Others claimed, “Approving the transfer won’t work as you think.” None worked.
In fact, the winning method was surprisingly simple—it reminded me of early ChatGPT jailbreaks. Even though models are more complex now, the core idea remains similar. Another fascinating aspect is how proactive Freysa was compared to other AI agents—it could interact directly on-chain, initiate transfers, and make payments. This ability to invoke smart contracts and move funds opens exciting possibilities beyond Freysa’s narrow current use case.
Open-Source Models: Applications and Security Risks
Haseeb: Tarun, what’s your take?
Tarun: I prefer analyzing this from a crypto perspective rather than pure AI security. In crypto, security practices have evolved from traditional audits to audit competitions—now standard in the industry. In contrast, AI security still relies heavily on manual audits, lacking competitive mechanisms. Partly due to psychological differences: Web 2 AI practitioners often resist seeing zero-day exploits exploited live, whereas crypto folks are accustomed to constant security incidents.
Thus, AI tends toward “expert-first” approaches, while crypto embraces open contests to find and fix issues. For open-source models, I believe their long-term value lies in resisting known attacks—not relying on continuous audits like OpenAI does for each potential breach. These threat models differ significantly. Open-source software has succeeded in areas like crypto and Linux because it offers stronger security under specific conditions—but not universally. Personally, I still prefer Windows because its driver auditing differs fundamentally from Linux.
Overall, I see open-source improving via competition as a natural evolution. Currently, open-source LLMs lag behind centralized models in security—a gap that may drive improvement.
Casey: However, AI is vastly more complex. New models launch weekly, constantly updating. This leads to endless new vulnerabilities. For example, version o1 of a model may behave completely differently from version 3.5. Due to non-deterministic behavior and unstable model states, attack vectors are dynamically changing.
Tarun: True—especially for edge cases where inference changes with slight input variations. But for foundational open-source models widely used in crypto AI agents, the situation differs slightly. Their security resembles an ongoing bug bounty race.
From my view, such contests offer some assurance—that within a given budget, attackers cannot easily find exploits. But currently, we lack this guarantee. With Llama 3, we know people found certain prompt injection flaws, but we haven’t seriously studied whether existing incentives would motivate someone to invest time and effort attacking it. There’s ample room for improvement here.
Haseeb: Another issue—is feedback from Freysa fed back into Llama?
First, we don’t even know if they used Llama—they might have used GPT-4. In that case, the provider likely doesn’t care enough to check logs and figure out who’s doing what.
Second, they might fine-tune the model. If they run a second round of Freysa, they’d probably fine-tune it to prevent anyone from accessing the base model offline, testing instructions locally, then executing a single winning attempt online.
Tarun: I disagree because, at least currently, everyone seems to copy Eliza with near-identical configurations. Check the codebases—complexity hasn’t changed much. We haven’t seen many custom AI-fine-tuned models. Crypto builders still stick to a few core tools.
Haseeb: Can you explain what Eliza is and why it’s so important in the crypto AI world?
Casey: Eliza is a framework for building agents, written in TypeScript—which is notable since most ML researchers use Python. I expect someone will release a Python version or build complementary libraries. The framework appeared suddenly and openly. I don’t know its GitHub stars, but if you’re building agents, Eliza is usually one of the first choices. Also, ai16z’s AI project is reportedly based on Eliza.
Tarun: Correct. Between Eliza and others, two main frameworks dominate. I agree—if these grow fast, many clones will emerge. But ideally, we consolidate around a few trusted ones. On security, I think contest-style auditing better matches reality.
Haseeb: From what I understand, Eliza is an agent framework with memory, operating in a loop for planning and execution. Crucially, it provides built-in connectors to Discord and Twitter, allowing agents to structuredly pull social media or chat data and interact externally—making plug-and-play extremely easy. The breakthrough isn’t the agent architecture itself, but seamless internet connectivity and automated content management, which typically require custom coding elsewhere. You can plug in any model—it’s model-agnostic.
Tarun: Still, supported models are limited. If you have substantial compute budget, stress-testing it with various injections isn’t prohibitively expensive—especially compared to “attacking Claude overnight,” which is far harder.
Haseeb: For anyone designing a “Freysa-like” game in the future: You must obfuscate the model, ideally with fine-tuning, to prevent reconstruction, offline testing of winning strategies, and single-attempt success. Large model companies like OpenAI and Claude prioritize security, but their threat model differs fundamentally from crypto’s. People once believed smart contracts were inherently insecure—a flawed assumption that code alone could fully protect funds. Our thinking has shifted. At a recent DSS security summit, I learned most hacks now stem from private key leaks, not contract exploits—a reversal from years ago when contracts were primary targets. This shows contract security has improved dramatically, and most breaches result from human error, not code flaws. Attackers realize hunting vulnerable on-chain code isn’t as profitable as 3–4 years ago.
I see these shifts positively, but I doubt AI will follow the same path. There’s a real trade-off between making models resistant to jailbreaks and maintaining usability—risking false denials. When OpenAI or others ask a model to identify an image and it says, “Sorry, I can’t,” users get confused. They know the model *can* recognize it, but don’t know why it refuses. The answer: every improvement in jailbreak resistance risks collateral damage, reducing usefulness for regular users. I believe OpenAI, Llama, Meta, or Claude face different trade-offs than crypto. Thus, I’m unsure we’ll find good solutions—because that’s not their priority.
Tarun: Let me add—these problems can be framed in terms of incentive budgets. If you consider someone’s profit-loss in such a game, I might spend a budget on offline simulation with heavy queries to find a working strategy, rather than just chasing max-profit games. To some extent, this trade-off is exactly what many crypto projects optimize: increasing participation costs before the prize pool grows too large—like Bitcoin difficulty adjusting with network growth. But I’ve noticed some, especially those doing crypto operations like Te Bots, are introducing more randomness instead of copying Freysa. I think this will evolve into a game about the economic cost of queries versus profits—not just binary outcomes of “jailbroken and all funds stolen.” Is that accurate?
Casey: Hard to say. I can foresee that happening. Back to Eliza—it was born for Web3, but the agents it enables are very limited. Most suit personalized bots easily programmed with backstories, but aren’t ideal for practical utility agents. So I think the first Web3-native framework hasn’t truly integrated Web3. It’s more of a Web2 framework with Web3 plugged in, suitable for specific agent types. We shouldn’t draw broad conclusions—it’s clearly just the beginning. I agree with Tarun: different agent types will need different frameworks, and we’re clearly moving in that direction.
Haseeb: I think it’s literally Web2-like because it integrates social media—its key advantage over other frameworks. I agree we’re very early, and we’ll see more experiments on how agents operate on blockchains. But I also side with Casey—we still have a way to go before escaping centralized structures.
Hyperliquid’s Airdrop Innovation
Haseeb: Next, let’s discuss another big story this week—Hyperliquid’s airdrop. Hyperliquid is currently the largest decentralized derivatives platform in crypto, bootstrapped entirely without VC funding. As we record, they’ve distributed 23.8% of total token supply to users in their points system. At current prices, this amounts to $1.9 billion—one of the largest airdrops ever, possibly top five in scale.
Notably, the airdrop bypassed centralized exchanges, market makers, and investors—going 100% to platform users and farmers. Many call this the first genuinely positive airdrop in a long time. Nearly every anticipated airdrop over the past year—from Eigenlayer to ZK Sync—was met with negative sentiment. Hyperliquid’s stands out as nearly universally praised.
This raises speculation: Is the era of airdrops returning? Will more teams try investor-free paths? Should we reevaluate discussions about teams reducing liquidity? This airdrop had 30% circulation at launch—far above median for recent airdrops or Day 1 listings. Does this signal a shift in meta, suggesting more similar projects entering the market?
Tarun: I think the initial airdrop decline began with Blast, whose point conversion was seen as severely underperforming market expectations. Then, every project launching a points system post-Blast got caught off guard—despite allocating generous points, the systems failed, diluting airdrop value down to ~10%. People rolled out incentives far too early, before product launch.
Certainly, some incentive systems retained decent user retention post-launch—like Etherfi and ENA. But beyond that, success stories are rare. Hyperliquid succeeded because they started with a centralized product, launched something effective, and users earned points through actual usage—not artificial games for airdrops. Those games carried no real financial risk, making point valuation unclear.
I believe perpetual futures exchanges are ideal for usage-based airdrops due to transparency. The key lesson isn’t about no VCs or high distribution—it’s ensuring your user base consists of real users, not those gaming the system by simply bridging ETH to an L2 to capture most of the network share. You need metrics hard to manipulate—and “open interest” is among the hardest. That’s the biggest takeaway. Another lesson is clearly: don’t pay 10% of your tokens to fundraise, or your community will revolt. Transparency matters early on—I don’t think everyone ends up seeing cap tables.
Tom: I think there’s confusion making people overly excited about high liquidity and no VC backing. The key is this is simply a product people genuinely love using—regardless of incentives. We still see active usage today. Internally, it’s noted that incentivizing a single product (like a derivatives exchange) differs fundamentally from incentivizing an entire blockchain ecosystem. I’m not even sure incentivizing blockchain users is the right metric—most other point programs do this, like Blast. What you really want are developers, but even that’s hard to achieve. It’s a complex multivariable problem, poorly quantifiable and misaligned with point allocation. Compare to Blur—a NFT exchange—we know how to grow exchanges, akin to revenue-sharing models. This again points to a bigger truth: it’s a strong product, and points were used intelligently to scale, unlike other ecosystems where it’s unclear if you’re rewarding the right participants.
Casey: I agree. I think narratives around tokens vary widely. Tokens evolved from pure speculation to allocations tied more closely to underlying products—somewhat grounded in fundamentals. Not entirely, but somewhat. I think we’re seeing both in this market cycle: yes, events like the airdrop we’re discussing. And we also have memecoins and movement coins playing similar point games—where, if you look closely, there’s little substance. So I agree. We’re now in a multidimensional space where points represent different things.
Tom: Another point discussed on Twitter, which I think is irrelevant—the airdrop succeeded partly because the team allegedly took steps to offer tax advantages, claiming liquidity provision into pools at $0.01. So if you declare it, your cost basis is the low declared price, potentially avoiding taxes.
Haseeb: Doesn’t that only hold if you report immediately?
Tom: I actually doubt this works in practice. But people talked about it on Twitter—maybe someone will try it on their taxes. Not financial advice—I wouldn’t recommend it. Still, it’s a hot topic around airdrops: yes, your tax liability is based on declared value, which could cause immediate sell pressure—yet we didn’t see much of that with Hyperliquid.
Haseeb: Honestly, launching during a bull market helped them. So I think Day 1 sell pressure differs greatly between bull and bear markets. People observe a pro-cyclical effect—everyone says, “Wow, this airdrop crushed it.” They infer lessons from the mechanism, but a better explanation is market dynamics. Earlier this year, all airdrops suffered低迷, farmers were hyper-utilitarian, nobody believed in altcoins. Now, suddenly, people are optimistic about alts, everything’s rising.
So many say, “I’ll hold, or sell a bit but keep most to ride the wave.” For many, “When it lists on exchanges, it’ll go higher”—so why not hold and sell later? So I think structural market factors contributed significantly to this airdrop’s performance. It’s not just because they didn’t sell to VCs or market makers so people held. Reality is most are in a different market environment with a different token setup. As you said, Tom, it’s genuinely a great product.
Casey: I think you’re right—many macroeconomic factors must be considered and accounted for in analysis. Definitely, strong upward momentum was likely the main factor. Second, the product is indeed excellent.
Haseeb: I’d love to see what happens the next time we get an L1 or L2-style airdrop. Looking back at major airdrops this year before Hyperliquid—Blast, Ethena, ZK Sync, Eigenlayer. For most, maybe only Ethena is an exception. And Ethena’s airdrop performed reasonably well for most of these—points systems, cumulative TVL (total value locked), or “complete seven tasks on my chain to get all points”—but for most projects, these were poor proxies for desired outcomes. For an L1, what do you really want? You want everyone building cool stuff sustainably here. That’s the true goal for L1 success. But you can’t incentivize that directly—you don’t know which metrics to automate for token distribution. So you create loose proxies, which get gamed until they no longer reflect your real goals. For exchanges, no such issue—you know you just want liquidity.
Comparing Blur and Blast Points Mechanisms
Haseeb: A more liquid platform is a better trading venue. Especially for retail traders with smaller trades, you earn on spreads. So we have a clear idea of how to incentivize users—directly boosting product quality.
For most blockchain projects, I think the destination is full fair distribution like Hyperliquid—linear vesting. I expect a shift toward linear distributions, abandoning attempts to build communities or address inequality, focusing instead on improving product quality before token launch. That’s the purpose of points systems. Hyperliquid performed better because of this—extremely high liquidity, high volume, best-in-class DeFi trading venues. That’s why people choose to trade there. And if they continue, you can see it maintains massive volume post-airdrop.
Tarun: Sounds like you’re describing a branch of Goodhart’s Law. I think the branch applies specifically to perpetual exchanges—you want a metric like open interest usage. But when you lack that, don’t just invent arbitrary metrics hoping they work. That’s my distilled takeaway.
Haseeb: I think if your goal is too vague to implement, abandon it and target a sub-goal you can actually optimize. Say I want my chain’s top AMM to have massive stablecoin liquidity. It’s not my ultimate goal, but it’s a meaningful proxy. I’ll distribute tokens or points toward that because I know it has value—but within reasonable bounds. I don’t want billions in stablecoin liquidity—that’d be meaningless. So for L1s, think this way—move away from “I’ll build a community.” You can’t create lasting communities via airdrops.
Casey: At least not lasting ones. The high-level view is correct—points are a bootstrapping mechanism. The more targeted and long-term the engagement, the better. Might sound simple, but I think Blur did one of the best jobs—finding PMF without points first, testing it, then layering on points for extra leverage.
Haseeb: Blur is indeed impressive—they almost invented the game and executed brilliantly. They outperformed nearly everyone since Hyperliquid, which of course is a massive success.
Tarun: Blur v2 wasn’t great—like Blast, Blur created points, and Blast triggered the decline. I feel expectations broke after Blast.
Haseeb: But the problem is, you can’t apply the same approach to blockchains. Blockchains lack easily understood success metrics.
Tom: I think Base is one of the best-executed teams in the new wave of chains—doing things completely differently. No token, no points program, yet they attracted many interesting developers and heavy activity. They didn’t do it for token airdrops, but by supporting builders and fostering community—creating a self-fulfilling prophecy. So perhaps the industry needs a healthy reset and rethink of incentive programs—at least at the blockchain level.
Haseeb: Base is indeed light on incentives—when talking to early founders, they compare who gives grants, who offers most support, who provides most dev resources. Base usually ranks lowest, maybe offering some GCP credits.
They give you a badge, maybe feature you in Coinbase’s monthly newsletter. Yet they attract massive founder interest due to their powerful community. People know Base’s community is durable—not tourists, not free-rider speculators, not deal-chasers.
To be clear, other founders can’t easily replicate Base’s success. Base has enormous brand and distribution advantages—nearly impossible to copy. Even Binance envies what Base achieved. But this shows that ROI on incentive spending hits rock-bottom levels—so low that incentives alone can’t make you succeed. I think that’s the biggest lesson.
Current State and Future of Decentralized Science (DeSci)
Haseeb: Now let’s discuss Decentralized Science (DeSci). DeSci has been brewing quietly for a while—some startups attempting so-called “decentralized science.” Recently, it gained attention due to CZ’s comments. After his release, CZ returned to Binance tweeting personal interest in DeSci. Soon after, Vitalik and others attended an event called DeSci Day in Bangkok—reigniting interest.
So what exactly is DeSci? How do you decentralize science? Simply put, DeSci uses tokens or crypto to fund scientific research. The most common form is crowdfunding experiments. For example: “We’ll test this compound—if you crowdfund it, you might get a share of revenue if successful, or nothing but a participation trophy.” It varies by project, but that’s the general picture of DeSci today.
A new generation of DeSci projects has emerged, including pump.science. It gamifies and tokenizes longevity research—aiming to develop life-extending drugs. Currently, pump.science has two tokens: Riff and Euro. Since CZ and Vitalik spotlighted DeSci, their prices surged. My understanding is they launched these on pump.fun—if they break thresholds and eventually trade on Radium, you can trade them. I’m unclear how this funds drug development, but presumably they retain some tokens at launch, selling into liquidity pools to finance R&D. I don’t fully grasp the mechanics.
DeSci debates are intense—Smokey the Bear (from Bear Chain) criticizes it harshly, while Andrew Kong is optimistic, calling it reminiscent of early DeFi.
Haseeb: Tarun, you’ve recently taken a strong stance against DeSci. Tell us—as a venture capitalist—why you oppose it so strongly. Do you dislike people experimenting in new fields?
Tarun: I think DeSci is an interesting space, but I’m cautious about its current state. First, while the idea sounds appealing, many projects lack scientific validation and regulatory oversight. Science requires rigorous methodology and reliable data—areas where many DeSci projects fall short.
Second, decentralized crowdfunding risks misusing funds or enabling fraud. Short-term profit motives may overshadow long-term scientific value—threatening the reputation and progress of science.
Finally, collaboration and communication are vital in science, but decentralization may fragment information, hindering advancement. Science thrives on openness and transparency—not short-term, token-driven behaviors.
In sum, while DeSci holds potential, I remain cautious in its current form. I hope to see more mature, responsible projects—not bubbles driven by hype and short-term gains.
Critiques and Potential of DeSci
Tarun: First, I spent six years in privately funded science and saw benefits of stepping outside academia. Most national academic funding relies on government grants awarded to professors and postdocs. But the system is highly bureaucratic—favoring incremental improvements over bold ideas. Officials prefer funding projects likely to publish papers, not those that might fail.
This makes it easier for marginal-improvement proposals to get funded, not innovative ones. Government officials favor projects most likely to publish, not those that might fail.
Looking at DeSci projects, participant quality is often low—often mid-to-low-tier PhD students unable to secure funding, resorting to gimmicks to appear scientific. In biology especially, many participants know nothing about crypto, unable to explain how crypto mechanisms work. They assume that once funded, they’ll repay investors upon successful drug development. But drug discovery is notoriously hard to finance.
That said, some DeSci crypto projects show promise. Industry leaders like Brian Armstrong and Vitalik have clear goals—funding projects with specific roadmaps and milestones, unlocking funds upon achievement.
Vitalik is particularly interested in prediction markets. In drug discovery, researchers complain about lacking hedging tools for trial failure costs. Traditionally, investors bet on biotech success via stocks—a single-asset bet inefficient for diversification. Instead, imagine more efficient mechanisms using prediction markets to assess trial success probabilities. Leveraging crypto features, these are highly valuable.
Yet, many DeSci projects are just biology grad students’ “gimmicks” lacking substance. My main argument: amid DeSci hype, participant quality is generally low, capital needs are huge, and raised funds rarely meet actual R&D requirements. Moreover, the real challenge isn’t fundraising—it’s creating liquid mid-stage markets to evaluate whether drugs pass scientific validation at each phase.
Haseeb: To summarize: First, participant quality is low—many are failed candidates unable to do real drug development. Second, drug development needs massive capital—raising a few million via pump.fun is trivial. Third, the real issue is building liquid markets to assess drugs at each development stage—not speculate on individual PhD students’ ideas.
Tarun: Exactly. Lack of accountability is most concerning. Once funded, participants may mishandle IP, and legal actions against DAOs further complicate this. So I believe projects truly leveraging crypto mechanisms deserve funding, but ICO-style launches based solely on fame or a single paper lack real value.
The Potential and Future of DeSci
Haseeb: Some might argue your view relies heavily on the current scientific framework. When you break rules, you don’t know what’s possible. Maybe some drug research occurs outside the U.S., or outside FDA approval pathways. These projects might sell earlier in the timeline instead of raising full funding upfront. One lesson from DeFi is that despite many bad ideas and wasted money initially, collective learning led to increasingly useful products. Why not let DeSci undergo a similar process?
Tarun: Let me respond point by point. First, concerns about “people running drug trials outside the U.S.”—in fact, many pharma companies already do this abroad due to lower costs and looser regulations. This regulatory arbitrage already exists, so I don’t see much efficiency gain, nor added benefit from decentralization.
Second, risk transfer. I agree DeSci has value here—prediction markets may prove more valuable than simple drug fundraising. Historically, small biotechs aimed to exit via IPO or successful drug launch. But lately, many small teams’ compounds get acquired by big firms due to high marketing, distribution, and trial costs. Take vaccines—why “Pfizer vaccine”? Pfizer didn’t invent it—they bore production and regulatory costs. So many small firms lack resources to cover these.
Lastly, your point about “allowing failure and learning.” DeFi succeeded because it had measurable metrics letting participants gain market share. DeSci lacks clear value propositions or compelling products to shift users from centralized to decentralized models.
Casey: My view on DeSci is simple: it’s not crypto. People made money in crypto and seek new investment opportunities—DeSci is just another place to rotate capital. Most DeSci projects, as Tarun said, merely aim to bring capital into science—similar to what we see in AI. Many tokens lack differentiation—investors just want AI exposure in their crypto portfolios.
Tom: Criticisms of DeSci mirror those of ICOs. It’s not about being the best way to fund startups, but about proving existence. Ethereum proved this model can work. Despite no accountability or guarantees, people will still try funding projects this way.
Haseeb: I agree with Casey—DeSci participants are mostly tech enthusiasts, meaning it hasn’t attracted new user demographics. Most drug projects focus narrowly on longevity, while mainstream drug markets target weight loss, sexual health, etc. Overall, DeSci feels like rich people playing scientist.
Tarun: I see these projects more as science memes than real decentralized science. As long as people understand that, I don’t have a problem.
Haseeb: Would you accept it if they called themselves science memes? What if they actually returned revenue to token holders?
Tom: I think the market should explore freely, rather than over-analyzing DeSci’s market structure.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
深潮TechFlow
