An AGI Moonshot, Not an LLM Factory: Hone’s singular focus is solving the ARC-AGI benchmark to achieve true generalization. This is a high-risk, high-reward play for a step-function leap in AI, not just another incremental improvement.
Architecture Over Data: The strategy is to out-innovate, not out-collect. By exploring novel architectures like JEPA, Hone aims to create models that think more efficiently and don't depend on ever-expanding datasets, sidestepping the data moat of centralized giants.
The Business Model is the Breakthrough: There is no immediate revenue. The investment thesis is straightforward: solve AGI, earn the ultimate bragging rights, and then monetize the world’s first truly intelligent model through distribution partners like Targon.
Vertical Integration is Non-Negotiable: To build AGI, the old model of horizontal specialization is dead. Owning the stack—from research to infrastructure to product—is the only way to move fast enough.
Ship to Socialize: Don't build AGI in a lab and drop it on an unsuspecting world. Products like Sora are deliberate steps to co-evolve technology with society, managing impact through iterative, public-facing releases.
The Real Turing Test is Science: The true measure of AI's power is its ability to make novel scientific discoveries. Altman believes GPT-5 is already approaching this milestone, which will have a more profound impact on humanity than any chatbot.
Stop Fearing Parameters. When in doubt, go bigger. Scale is not just about capacity; it’s a tool for inducing a powerful simplicity bias that improves generalization and paradoxically reduces overfitting.
Trade Hard Constraints for Soft Biases. Instead of rigidly constraining your model architecture, use gentle encouragements. An expressive model with a soft simplicity bias will find the simple solution if the data supports it, while retaining the flexibility to capture true complexity.
Think Like a Bayesian. Even if you don't run complex MCMC, adopt the core principle of marginalization. Techniques like ensembling or stochastic weight averaging approximate the benefits of considering multiple solutions, leading to more robust and generalizable models.
Reward Function is Everything. Mantis’s success hinges on its information-gain-based reward system, which attributes value based on a miner’s marginal contribution to a collective ensemble, not just their individual accuracy.
Inherent Sybil Resistance. By rewarding unique signals, the incentive mechanism naturally discourages miners from running the same model across many UIDs, solving a critical vulnerability in decentralized AI networks.
The Product is Verifiable Alpha. The endgame is not just to build a subnet but to produce a monetizable product: high-quality financial signals, auctioned to the highest bidder and backed by an immutable on-chain performance record.
Incentives Dictate Intelligence. Mantis's breakthrough is its reward function. By precisely measuring a miner's marginal contribution, it makes unique alpha the only profitable strategy and naturally defends against Sybil attacks.
The Ensemble is the Alpha. The network’s power lies not in finding one genius quant, but in combining many good-enough signals into one great one. The collective intelligence is designed to be far more valuable than any individual participant.
The Future is Verifiable, On-Chain Alpha. Mantis plans to monetize by auctioning its predictive signals, creating a transparent marketplace for intelligence and proving that a decentralized network can produce a product valuable enough to compete with Wall Street's top firms.
Google's "Tax on GDP" Is Under Threat. AI is eroding the informational searches that feed Google's funnel and will eventually intercept high-intent commercial queries, redirecting economic power to new agentic platforms.
The Future of Shopping Is Agentic, Not Search-Based. Consumers will delegate research and purchasing to specialized AI agents that optimize every variable, from product choice to payment method, fundamentally changing how brands acquire customers.
Trust Is the Ultimate Moat. In a world of automated "crap," business models built on human trust and strict curation, like Costco's, become exceptionally defensible.
AI's next frontier isn't just language; it's simulating life. The "virtual cell"—a model that predicts how to change a cell's state—is the industry's next "AlphaFold moment," aiming to compress drug discovery from years of lab work into forward passes of a neural network.
Biology's core bottleneck is physical, not digital. Unlike pure software, progress is gated by the "lab-in-the-loop" reality: every AI prediction must be validated by slow, expensive physical experiments. Solving this requires new platforms that can scale the generation of high-quality biological data.
The biotech business model needs a new playbook. With a 90% clinical trial failure rate, the economics are broken. The future belongs to companies that either A) use AI to drastically improve the hit rate of drug targets or B) tackle massive markets like obesity, where GLP-1s proved the prize is worth the squeeze.
Enterprise AI is a Services Business. The best models are not enough. Success requires deep integration via "Forward Deployed Engineers" who build the necessary data scaffolding and orchestration layers.
GPT-5 Was Co-developed with Customers. Its focus on "craft" (behavior, tone) over raw benchmarks was a direct result of an intensive feedback loop with enterprise partners, making it more practical for real-world use.
Bet on Applications, Not Tooling. The speakers are short the entire category of AI tooling (frameworks, vector DBs), arguing the underlying tech stack is evolving too rapidly. Long-term value will accrue to those building applications in high-impact sectors like healthcare.
Intelligence Has a Size Limit: Forget galaxy-spanning superintelligences. The physics of self-organizing systems suggest intelligence thrives at a specific scale, unable to exist when systems become too large or too small.
True Agency is Self-Inference: The crucial leap to higher intelligence is not just modeling the world, but modeling yourself as a cause within it. This recursive "strange loop" is the foundation of planning and agentic behavior.
Hardware is the Software: Consciousness is not an algorithm you can run on any machine. It likely requires a specific physical substrate where memory and processing are unified, making the body and brain inseparable from the mind.
**Value is a Function of Time:** Bitcoin's greatest asset is its 15-year track record. Lasting value isn't about technology alone; it's about a powerful story that withstands the test of time, creating an insulated brand.
**Self-Custody is the Premise:** The entire value proposition of crypto hinges on eliminating counterparty risk. Compromising on self-custody and security for the sake of convenience is a recurring mistake that "always blows up."
**Adoption Will Be Abstracted:** The future of crypto for the masses is one where the complexity is hidden. Centralized user experiences will run on decentralized rails, delivering the benefits of crypto (lower fees, faster settlement) without the unforgiving user experience.
**Stop Gambling, Start Engineering.** The biggest edge isn’t in predicting price but in finding and exploiting structural market inefficiencies. Focus on trades where you can control or heavily influence the outcome, like RFV plays or creating self-fulfilling prophecies in prediction markets.
**Become the Casino.** The crypto market is filled with speculation. By providing liquidity, farming yields, and taking the other side of gamblers (e.g., selling Pendle PTs), you can generate consistent, lower-risk returns. Farmers, on average, outperform directional traders over the long term.
**Alpha Lives in the Weeds.** The most significant opportunities aren’t on the front page of Twitter. They’re buried in obscure Discord servers, complex protocol mechanics (like Aerodrome’s bribes), and emerging platforms with low capital efficiency like Polymarket.
Private Markets Are the New Public: The real unlock for tokenization isn't just 24/7 stock trading—it's bringing high-growth private companies to retail investors, with or without the company's blessing.
The Great Convergence Is Here: The line between a crypto exchange and a stock brokerage is disappearing. Robinhood and its competitors are converging on a single "financial super app" model where all assets live in one place.
Regulation Has Created a Paradox: The current system allows unlimited speculation on assets with zero fundamental value (memecoins) but blocks access to premier private equity. Robinhood is betting this logic won't hold.
Embrace the Friction: The current difficulty of investing in Bittensor subnets is a feature, not a bug. It’s the moat that has suppressed valuations, creating an opportunity akin to buying Bitcoin on Mt. Gox before Coinbase existed.
A 3-6 Month Catalyst Window: The development of bridges and institutional infrastructure is the primary catalyst. This window represents the final moments to gain exposure before capital can flow in easily, likely re-rating the entire ecosystem.
Think Startups, Not Just Tokens: Evaluate subnets like early-stage companies. Use resources like the *Revenue Search* podcast to analyze financials and projects like Shush (AI inference), Score (AI vision), and Quantum (public quantum computing) as real, venture-style bets.
**Don't Panic Sell.** The current market dip is a sign of a healthy "wall of worry," not a cycle top. Historical on-chain indicators show there is significant room to run.
**Follow the Smart Money.** Institutions are aggressively buying this dip. The real capital from pensions and sovereign wealth funds is still on the sidelines, waiting to enter.
**The Fed is Turning Bullish.** A key Federal Reserve official is now openly advocating for crypto adoption within the regulatory apparatus, signaling a major long-term shift in the US.
**The Dollar Isn't Being Debased; It's Deflationary.** The market is not pricing in inflation or debasement. Instead, key indicators like the interest rate swap market are emphatically signaling a future of much lower interest rates for much longer, which is characteristic of deflationary pressure and a strong dollar.
**Asset Booms Are a Symptom, Not a Solution.** Rising stock and crypto prices are not evidence of a healthy economy or money printing. They reflect a K-shaped recovery where capital flees into financial assets as a hedge against systemic fragility, while the real economy for labor remains stagnant.
**The Contrarian Play Is Long Bonds.** If the global system is starved for safe, liquid collateral and headed toward a deflationary recession, the best-performing assets will be long-duration U.S. Treasuries. Snyder’s advice is the polar opposite of the typical crypto portfolio: be long bonds.
