
Things I wish I could do with AI (that is still surprisingly hard) 🧵

Senior Product Director @zama. Co-founder @cheqd_io. Co-chair of Technical SteerCo @DecentralizedID. Ex @FinTechLabLDN, @inside_r3, @Accenture

Things I wish I could do with AI (that is still surprisingly hard) 🧵

If you’re building vaults or staking into a product, @yield_xyz is the de-facto. And now, it’s easier than ever to bring this to your own product with support for confidentiality.
Confidential yield is coming to more of the apps people already use. @yield_xyz partners with Zama to bring confidential vaults into its Earn Widget and API. Wallets, custodians, neobanks, and fintech platforms can now offer confidential yield within their existing Earn products. For users, that means more ways to earn onchain through the apps they already use, without exposing their individual positions to the public. Read the announcement: zama.org/post/yield-xyz-inte…

Kinda interesting geopolitically that the Jensen Huang story broke a week or so before Anthropic IPO. If it was gonna come out at some point, what better time for China to detonate this news to cripple key US AI competition and hope that local chipmakers like Huawei can last it out to catch up. 🧐

Cool use case but perhaps giving AI the precedent of stabbing things in the brain is not the best idea? 🙊
we have a winner for the most absurd AI startup ever this Peter Thiel backed company uses AI robots to stab fish in the brain so they taste better and last up to 3x longer it's called Shinkei and it modernizes a centuries-old Japanese technique dozens of Michelin-star restaurants are already lining up to buy the fish here's how it works: > the problem is when a fish gets caught, it usually just suffocates on the boat > that stress floods the meat with hormones and lactic acid, which hurts the flavor and texture > the old Japanese fix for this was spiking the fish through the brain right after the catch, which stops all that > but doing it by hand is impractical for most American fishing crews > so Shinkei built a fridge-sized robot that sits right on the boat's deck > the crew feeds in a fish and AI identifies the exact species, so it knows exactly where the brain and gills are > the robot then drives the spike through the brain and cuts near the gills in about 7 seconds > the fish goes into an ice slurry so the blood drains out > so the fish doesn't sit there stressing out, and that tasty flavor and texture is preserved example number 5,392,417 of how AI can be applied to literally anything

I like the Michelin Guide comparison for the new ETHBeat interface, as it's an opinionated take outside of just soundness/security. Fun thought experiment with @donnoh_eth today: Michelin has the concept of a "Bib Gourmand" for a simple/good cooking, which is not "fancy". So if we're extending the Michelin analogy, what's the blockchain equivalent of that?
ETHBEAT Michelin guide to projects serving you best CROPS in town ethbeat.org/crops

“Encrypted, not anonymous”, a workshop by @OpenZeppelin today on how to build confidential contracts using @zama FHE, and common types of gotchas to look out for to build secure contracts. Also got to meet companies wanting to use confidential contracts in @lfdecentralized blockchain stacks!

Great to see the crew at Devcon and supporting @lfdecentralized booth. How time flies. In 2019 Consensys contributed 🙏 Pantheon, renamed to Besu and now it's one of them most used instituional protocols. 👀 Institutional Adoption report for more details lfdecentralizedtrust.org/led…
Besu started at Devcon in 2018 under the name Pantheon. Eight years later, the Consensys teams behind @Besu_eth and @Teku_Consensys are back at Devcon, this time in Mumbai. Whether you’re already running an institutional Besu ledger or looking to build one, run a node, build on Ethereum or contribute, come find us this week. @EFDevcon @mkriak

Imagine the end game here is that DPRK just holds the ETH and waits for stablecoins to come back + EEZ and then they rug the network by starting to censor the network via becoming the largest ETH validator network and conduct a sustained 51% attack
This is a great meme, but it's really not a joke. This is creating a real risk for Ethereum. If DPRK hacks continue in the same trajectory - DRPK can be a meaningful shareholder in Ethereum. How does that play with TradFi? 🧵

This is a great meme, but it's really not a joke. This is creating a real risk for Ethereum. If DPRK hacks continue in the same trajectory - DRPK can be a meaningful shareholder in Ethereum. How does that play with TradFi? 🧵
DPRK holds approximately 420.000 ETH from the Binance hack Ethereum Foundation has approx 170.000 ETH in its treasury DPRK:

Echoes of this in the talk today from @Elliot0x: we need a lot more timelocks to fix / undo hacks, as exploits become more common. At least as a choice, if not a default.
This takes me back to when I was working for banks on core banking related systems. The security process they had in place were extremely defensive and we need more of that thinking in the crypto industry: - No engineers, ever, had long-lived elevated or write permissions. Elevated permissions would only be given for very short windows using TACACS/RADIUS etc. Modern DevOps layers many layers of independent MFA but fundamentally trusts the person. Bank policies would not assume that trust by default. - Little to no automation, and then each granular change step checked four-eyes. This slowness/speed annoyed me a lot (“why can’t we one-shot this in minutes with automation!”) but the goal I understand now was to deliberately slow everything down so that no change was rushed. - Everything would be scheduled to happen within a change window of a few hours max., with half the time or more dedicated for backing out the change if it didn’t succeed. - Insane levels of circuit breakers and cutover redundancy in the infrastructure to fail over fast to working/non-compromised systems. - “Defense in depth” of logging, auditing, and monitoring tools, where there would be multiple tools from independent software vendors monitoring and auditing the same layer. Modern DevOps shuns a lot of this (not unusual to have a single monitoring/auditing system at a layer, e.g., “just whatever AWS has”) - A *lot* more focus on “human factors” like is the change at a time when people are at peak focus (or get enough rest before), too many engineers on holiday (as backup if things escalate). Everything had a Plan B, C, D, E. We were always asked to think of the headlines in the newspapers the next day if a new feature/rollout went wrong: “if this rollout goes wrong, this is how much the UK GDP would tank by tomorrow”. Many of these “slow down” policies frustrated me at the time when fintechs/neobanks were shipping faster and lapping traditional banks. But there are lessons and practices in there which will become increasingly relevant for crypto.

What's the hottest-running local LLM that can be used to dry a spill on a t-shirt? Asking for a friend 🙊

It’d be funny someone made a privacy-preserving Guess Who? style game called DevKaun? at the @ETHGlobal hackathon later this week 😂

A developer with a face hidden by a mask would be “Dev kaun?” in Hindi (“who is the developer?”, pronounced similar to DevCon). I wonder if @VitalikButerin chose this outfit to go with the pun on purpose 😬
Vitalik on the main stage at Devcon Mumbai

If you opt of having your site data being used to train AI with Google, you also opt out of the site showing up in Google Search. Consequently, most websites don’t. It’s a scandal that Google’s use of market dominance to tie search and AI features together has not already been investigated by competition regulators. This will do far more good than any “AI Act”.
google says if you dont want AI in the classroom you wont get search either earnest Q: highlighted quote seems carefully crafted to suggest there is a technical reason google could not turn off AI in search. is there *any* truth to that? nytimes.com/2026/11/03/us/go…

First day at @EFDevcon Mumbai 🇮🇳. Impressions: - Feels like fewer people than in previous years. Market conditions? Visa difficulties? Maybe it's also unfair to compare with last year's Devconnect in Buenos Aires, which was absolutely massive 🇦🇷 - Lots of local talent: the Indian developer community is young, large and hungry - Agents are the big topic of the moment. Lots of talks, and walking down the halls you hear the word "agent" all the time - Relatively less talk about prediction markets than last year. Agents stole the spotlight - Privacy and zero-knowledge are still central - Very little institutional presence: maybe it has to do with restrictive Indian crypto regulations. It feels like a more cypherpunkish Devcon (a bit similar to Prague's Devcon 2018). The fact we’re in India also gives some “infinite garden” vibes - Quantum computing also shows up on the agenda - We're at the convergence of AI and quantum. I guess we're about to see what happens when massive data meets a new paradigm of computing. Is the Singularity right around the corner? - Lots of hallway talk about the recent hacks. The feeling is that we're entering the "big hacks era" with the advent of more powerful AI models - In this context, AI alignment becomes more relevant. As agents become more capable, they also become more dangerous, as the Hugging Face incident showed - Which is why my favorite talk was Civic AI, by @audreyt with @VitalikButerin. AI shouldn't be aligned only inside a handful of labs, but with the communities that use it - No idea why Vitalik is wearing a mask all day 🤣 - The chai at the coffee breaks is an upgrade over any conference coffee ☕ See you tomorrow for day 2!

Cue everyone saying I'm Spartacus! ... My point was it's not solved at scale. One one chain, or by one startup without distribution is not yet it
The problem with agentic commerce and personal agents is we're thinking about per task intent rather than broader scopes of delegation. Is anyone working on actually delegated authority? That's apparent