thomas

@balajis.eth was on stage today at SuperAI in Singapore. He shared several thoughts on how AI and crypto can complement each other (he also mentioned Farcaster a couple of times). I took notes and elaborated on them a bit in this 🧵:

PhD in aviation ∩ climate science | Aviator, diver, ETH maxi, nexialist | Host of /diving /ethfinance /eth-staker /geopolitics /science /singapore /thomas

1/ AI is probabilistic by nature (returning the highest likelihood output). Crypto is deterministic (following consensus rules, executing EVM opcodes, etc).

3/ AI is making content easy to fake. Crypto makes it hard again through the use of cryptographic signatures and onchain auditability of the entire history of the digital asset.

4/ Related to 3/, AI generates content while crypto authenticates it. One example is a Leica camera that uses Adobe's Content Authenticity Initiative to cryptographically sign the origin of a photo.

5/ AI weakens defense mechanisms against Sybils and bots; for example, it can now solve virtually all CAPTCHAs. Crypto rebuild those defenses (e.g., Sign In With Ethereum).

6/ AI proselytizes; it can craft compelling narratives that rewrite history in a particular way. With crypto, it is possible to point to specific blocks in an explorer (such as the FTX transactions on Etherscan) to prove some event happened. History becomes immutable.

7/ AI is breaking the public web by filling it with hallucinatory and zero-calorie garbage. Crypto solves that by enabling a verifiable web of trust and social reputation / ranking.

8/ Virtually all AI model training and deployment today is either fully centralized (OpenAI) or partly centralized (Llama). Crypto excels at crowdfunding projects that can help decentralize both training and deployment of open-source models.

9/ AI evaluation is centralized because it requires expensive GPU clusters. Crypto can decentralize the infrastructure (individuals get paid for selling their GPU time to virtual clusters that are permissionlessly available to anyone on demand).

10/ Centralized AI models cannot train on gated data (such as all scientific journals, or Disney movies). They also get weaker over time as IP infringement lawsuits erode their training set. Crypto doesn’t play by the rules and can train models on Sci-Hub, pirated content, etc.

11/ AI creates new authoritative sources of truth (such as ChatGPT). Crypto challenges them by democratizing multiple models aligned with different values and thinking. Analogous to monotheistic vs polytheistic religions. 🎬

2/ AI provides digital abundance; past the model training stage, prompting cost will eventually trend to zero. Crypto brings digital scarcity; not only artificially (fixed token supply) but also cryptographically, through non-fungibility and provable ownership (e.g., users can transfer but not copy digital assets).