In a current X publish, Ethereum co-creator Vitalik Buterin revealed he was doing a little bit of a self-experiment.
The objective of the experiment was to make use of his private well being and journey information to supply personalised food regimen and train suggestions for himself, utilizing frontier fashions however in a method that avoids leaking any personal info to them.
For the experiment technique, Buterin used an area mannequin (Qwen 3.8 Flash Subsequent) to orchestrate and in addition use highly effective distant fashions by way of device calls to profit from higher-level considering and information that the native mannequin lacks.
There’s a three-layer method to privateness: first, to keep away from leaking personally identifiable info, or leaking his identification by his writing model, because the native mannequin writes the queries to the frontier mannequin.
The second layer makes use of zkAPI to stop his identification from being uncovered by the cost channel, whereas the third makes use of Tor to supply community and IP-level privateness. “You want all three,” Buterin wrote.
Doing a little bit of a self-experiment.Purpose: use my private well being and journey information to supply personalised food regimen and train suggestions for me, utilizing frontier fashions however in a method that avoids leaking to them any personal info.Technique: use an area mannequin (Qwen 3.8 Flash Subsequent) to orchestrate, and use highly effective distant fashions as a device name to profit from higher-level considering and information that the native mannequin doesn’t have.Three-layer method to privateness:Keep away from leaking personally identifiable info, or leaking my identification by writing model -> native mannequin writes the queries to the frontier mannequin, not me Keep away from leaking who I’m by the cost channel -> zkAPI Keep away from leaking who I’m by networking / IP -> TorYou want all three (and at last we’ve all three, at the very least to some extent)A talent file teaches the native mannequin when and the way to assemble minimally-data-revealing requests to distant fashions. Use zkAPI-wrapped-with-Tor as a CLI device.And every thing works! I bought the suggestions again, data from frontier fashions helped to enhance them.Foremost deficiencies:* Tor is absolutely not optimized for request-by-request de-linking, which is the one type of network-layer privateness that actually is sensible in at the moment’s world (long-running identifiers are too fragile). Most likely not personal sufficient, and latency 10-100x increased than it might be, on the similar time. * The talent file’s request building methods are positively removed from optimum. * Qwen 3.8 Flash Subsequent continues to be too gradual for consolation. It is comfortably working at 20-30 TPS, however it will solely actually begin to really feel quick at 100+ * There’s a tradeoff: the extra cautious you might be about what information you give to a distant mannequin, the much less it will probably assist youhttps://t.co/1g81Mp1sWW
A talent file teaches the native mannequin when and the way to assemble minimally data-revealing requests to distant fashions. Buterin stated he used zkAPI wrapped with Tor as a command-line device.
In accordance with Buterin, the experiment labored and returned suggestions that had been improved utilizing info from the frontier fashions: “And every thing works! I bought the suggestions again, data from frontier fashions helped to enhance them.”
Buterin outlined three principal deficiencies, highlighting the trade-off; the extra cautious you might be about what information you give to a distant mannequin, the much less it will probably assist you to.
Ethereum imaginative and prescient stays
Whereas Buterin’s newest discourse on X seems to focus solely on AI, his imaginative and prescient for Ethereum stays unchanged, with him envisioning it as a ‘cryptographic world laptop’ within the years forward.
In a publish in direction of September’s shut, Buterin famous that Ethereum’s expertise has advanced enormously over the past fifteen years, and is poised to evolve much more over the subsequent three — to the purpose the place it turns into honest to name what Ethereum is shifting in direction of a qualitatively totally different form of system.
In accordance with Buterin, subsequent 12 months’s Pectra laborious fork would possibly possible be Ethereum’s final “regular” fork, with options and expertise that may be recognizable to somebody in 2015. The whole lot after that entails recursive STARKs, automated formal verification, extremely optimized consensus algorithms, and making all of it quantum-safe. With PeerDAS, Ethereum started its transition from being only a blockchain to being one thing far more highly effective. Beginning after Pectra, this transformation turns into Ethereum’s major story.
The end result is less expensive, extra scalable, and extra personal high-security computation than something that might be finished with the earlier period’s expertise alone: the cryptographic world laptop.
