Happy Sunday, friends!

As an audio learner, I’m hooked on podcasts and videos—I easily rack up 10+ hours each week listening while on the move or at the gym.

So, I thought, why not share my top takeaways from my favourite listens with you on the weekends?

Time is money, and this will save you plenty of both.

This weekend, we’re covering:

1. Are we overhyping AI in Crypto by Casey Caruso (Topology VC) & Mohamed Baioumy (EXO)

2. The Google x Bittensor presentations by JJ and Mog

1. Are We Overhyping AI in Crypto? (Empire)

• We’re witnessing a pivotal shift away from creating foundational models, and towards focusing on applications, infrastructure and agents. Large language models (LLMs) are quickly being commoditized.

• Finding product-market fit in the crypto AI space is difficult. Many products are not fully functional or scalable yet.

• For Crypto AI to succeed, it must deliver practical advantages like cost savings or efficiency, rather than relying on ideology alone.

Why Crypto?

• Crypto’s true power lies in its ability to coordinate and incentivize at scale

• AI is reaching the upper limits of publicly available data for training. This has sparked growing interest in how crypto could incentivize the creation and curation of new data sources.

• Projects like EXO focus on edge compute to allow businesses to run AI models privately on-premise. This can be particularly valuable for sectors with strict privacy requirements (e.g., law firms, and biotech).

• Advice for investors and builders? Investment in infrastructure (like GPUs or distributed compute) has been historically profitable, but there’s a push to think beyond copying past trends. Focus on solving real problems and not get too caught up in hype. Practical solutions will drive value.

Underrated and Overrated Areas in Crypto AI

Overrated

1. Privacy-focused products as a standalone feature: Privacy is more of a checkbox for regulatory compliance than a primary business driver.

2. Building new blockchains specifically for AI agents

Underrated

1. Hardware acceleration in the scaling of ZKML (Zero-Knowledge Machine Learning) and other verified compute applications

2. Crypto’s potential to streamline payments for AI agents

2. Google x Bittensor: Mogmachine

• Bittensor is essentially distributed compute at scale.

• Subnets within Bittensor are like individual startups that operate independently. Each of the 52 subnets offers different capabilities: inference, data processing, and specialized computations like protein folding.

• Subnets are evaluated based on performance, and those that provide higher value receive better rewards.

On Incentive Mechanisms in Bittensor

• Incentive mechanisms are crucial for attracting participants to subnets and ensuring they perform tasks efficiently.

• Dynamic incentive mechanisms are necessary to prevent stagnation and promote growth. Effective incentive mechanisms should not just reward participation but also quality and consistency of contributions.

• Eg. weighting scores based on criteria like timeliness, engagement, and overall performance

• There is a risk of participants “gaming” the system by exploiting predictable incentive structures. To counter this, it is crucial to ensure diverse sources of evaluation and continuous updates to the scoring criteria.

• The transition from synthetic tests (predefined evaluation criteria) to organic data (real-world usage) introduces complexity in designing fair and effective incentives.

On Trust and Decentralization

• Emissions are currently determined by validators, which can introduce biases. Bittensor aims to shift toward a market-driven system where value is assigned dynamically based on demand.

• dTAO (dynamic TAO) will allow participants to determine value in a more decentralized manner.

Some Tips for Miners & Subnet Owners

• Learn from others' mistakes by thoroughly studying the code base rather than relying on standard practices.

• Focus on leveraging Bittensor to develop practical, high-quality products, rather than just price prediction. Let’s be honest: if you could predict BTC prices precisely, you’d likely be running a hedge fund rather than working on Bittensor.

• Bittensor allows anyone to participate. If you have a strong idea, you won’t need to fund your own GPUs. Bittensor will do so.

3. Google x Bittensor: Joseph Jacks

On Open Source Software

• “Open source is eating software faster than software is eating the world”. Open source is advancing rapidly and outpacing the traditional software industry in many areas. This trend is driven by developers seeking open alternatives to popular proprietary software.

• Shift in Software Industry: We’re seeing the evolution from traditional software distribution models to SaaS, and now to commercial open-source software

• Commercial open-source companies operate under a “loose coupling” model where the core technology is open and freely accessible, allowing external developers to contribute and monetize the technology.

• Open-source businesses face unique challenges in monetizing free software. The business model often involves offering paid versions or complementary services.

On Bittensor

• JJ’s core thesis for Bittensor is “What if anyone without the permission of any government, corporation, society or entity could contribute to and own the most capable AI that humanity could collectively create?”

• Bittensor has no centralized ownership or pre-mine, distinguishing it from many other crypto projects. The founders hold <1% of the tokens.

• Bittensor is incentivising the continued production of intelligence (similar to how Bitcoin is doing that for money)

• His advice for TAO? Buy, hold it for 5 - 7 years and don’t trade, as it will be very volatile.

• Introducing a stablecoin within the ecosystem could help mitigate this volatility.

That’s it! Saved you 2+ hours of time. Have a wonderful rest of the weekend.

Cheers,

Teng Yan