Ho Chi Minh City, Vietnam has been fun, but now it’s time to meet a few of our developers in Hanoi and get our core technology tested to death ☠️

More team members of @KRNL_xyz getting the latest alpha at @Convictionvn

By sheer coincidence I found out @eth_vietnam is on at the same time as @Convictionvn in HCMC so I obviously had to attend. I got to meet the founder and connect with her so I’m even more excited.

Just landed in Ho Chi Mi City, Vietnam to attend the Conviction 2025 event:

Ask how it fails before you ask how it scales.

Fast iteration beats perfect planning.

Verifiable compute is coming faster than people think:. → zkML for inference. → zkTLS for data. → TEEs for secure enclaves. → Onchain attestations for provenance. It’s happening layer by layer.

The goal is not to “move the internet onchain.”. It’s to make the internet provable. AI, compute, routing, data, all of it. Composable and verifiable.

We’re entering the post trust phase of AI. It’s not enough to ask “what did the model say?”.You need to know:. ~ Which model?.~ Which weights?.~Which dataset?.~ Who approved the output?.~ Can I verify all of that cryptographically?. If not, it’s just vibes.

Crypto infra is now a public good for the AI economy:. ~ Global consensus.~ Neutral settlement.~ Cryptographic audits. We spent a decade building this. Time to use it.

Every infra cycle has the same story:. → More automation. → More delegation. → Then someone asks: “Who’s auditing the automator?”. That’s where verifiability comes in.

AI inference will get cheaper. AI infrastructure will get faster. But AI auditability will become mandatory.

Compute is cheap and getting cheaper. Trust isn’t. Verifiability will become the scarce asset in AI infrastructure.

Distributed compute got its execution layer (Rollups). Now distributed AI gets its provenance layer (KRNL). Verifiability is how real-world AI earns trust imo.

Next stop a short hop to Barcelona, Spain from Manchester, U.K. to attend the Tech Barcelona Summer Party 2025. #InBarcelonaWeTech

Observability ≠ explainability ≠ verifiability. Your dashboard might show the output. Your LLM might say why. But only a verifiable log can prove it, to another system, to a user, or to a regulator.

Snapshots are dead. Modern AI retrains weekly, updates weights daily, and serves new prompts hourly. If you’re not logging + hashing + fingerprinting every event at runtime, you’re flying blind.

It’s such a pleasure to welcome my co-founder @asim_eth to my home in the U.K. especially when he comes bearing gifts 🍦

Thank you to Apple for showing my nephew the power of the Vision Pro. I got to discuss the new features with the most recent update so I can take advantage of them on mine that I bought in February 2024.

I think Verifiable AI is base layer infra for regulated AI. You wouldn’t deploy finance without audit logs. Don’t deploy AI without cryptographic traceability. :).