Exchange Ranking
Top Exchanges
24h Volume Ranking
Popularity Ranking
Exchange BTC Balance
Proof of Reserves
Decentralized Exchanges
Funding Rate
Funding Heatmap
Liquidation Data
Max Pain
Long/Short Ratio
Whale L/S Ratio
Binance/Okex/Huobi L/S
Bitfinex Margin L/S
ETF Tracker
Solana ETF
XRP ETF
Hong Kong ETF
Bitcoin Treasuries
Crypto Reversal
Ethereum Reserves
HyperLiquid Wallet Analysis
Hyperliquid Whale Watch
Large Transactions
On-chain Movement
Bitcoin ROI
Stablecoin Market Cap
Options Analysis
News
Articles
Economic Calendar
Features
Wallet
Contract Calculator
Security
Collections
Watchlist
Following
ETH Key Insights
Vitalik Buterin combines native artificial intelligence, zkAPI and Tor networks to limit the exposure of personal data. The processing speed of the local artificial intelligence is 20 - 30 tokens per second, which is lower than his expectations of more than 100 tokens per second. Tighter privacy protections limit the contextual information remote AI models can use to provide advice.
Vitalik Buterin outlines three privacy levels for an experiment using personal health and travel data. His goal is to generate personalized diet and exercise recommendations while limiting the private information shared with powerful remote models.
The system uses local models, zkAPI payments, and Tor networks to address the different ways in which services may recognize him. Each aspect handles separate sources of exposure in the same workflow. Local model creation requests, zkAPI solves payment identification issues, and Tor targets network information.
Vitalik Buterin stated that the system returned recommendations and the information contributed by the remote model improved the results. Buterin also described shortcomings related to speed, request construction and privacy. These shortcomings include the trade-off between disclosing personal information and providing enough data to remote models to help improve.
Vitalik Buterin uses native AI to prepare requests
The first layer is built on top of the local model Vitalik Buterin identified as Qwen 3.8 Flash Next. Instead of sending his own phrases directly, he used the model to write queries for remote systems.
Vitalik Buterin 在他的 AI 实验中概述了三个隐私层级 / 来源：X
这种方法旨在防止泄露个人细节及其写作中的识别模式。技能文件指导本地模型何时咨询远程模型以及如何最大限度地减少披露的信息。这种安排允许实验利用超出本地模型知识和推理能力之外的功能。同时，本地系统作为更广泛推荐流程的一部分协调这些请求。
Vitalik Buterin 表示，远程模型有助于改善建议，但他发现请求策略远非最优。他还报告称本地处理速度约为每秒 20–30 个 token。为了获得流畅的体验，他希望速度超过每秒 100 个 token。
zkAPI 和 Tor 解决支付和网络身份问题
第二层使用 zkAPI 通过支付验证地址以访问远程模型。相关文档解释说，zkAPI 接受 ETH 或 USDC 存款以覆盖使用成本。此支付选项将测试路由到以太坊加密网络，而人工智能模型则处理规定的程序。然而，仅靠支付隐私并不能解决实验中描述的每个识别渠道。
As a result, Buterin added Tor as a third layer to protect information related to networks and IP addresses. He accesses the combined zk API and Tor settings through command-line tools. In summary, these layers address issues of request content, payment information, and network identity. Vitalik Buterin emphasized that the experiment requires all three safeguards. Still, his description acknowledges that their availability does not mean that every component provides sufficient protection or performance.
Privacy restrictions reduce speed and available context
Tor poses specific difficulties because Buterin says it has poor support for separation of individual requests. He questioned whether its privacy protections were sufficient to achieve this purpose. He also described latency as 10 - 100 times higher than a direct connection. This comparison focuses on potential performance issues rather than comparisons with explicit measurements of direct connections.
In addition to network latency, the processing speed of the local model made the workflow slower than he wanted. According to him, request construction also needs to be improved. In addition to these technical issues, the experiment also exposed limitations on the information available to remote models. More careful sharing allows these systems to have less context when generating useful suggestions. Buterin reported on a working setup, while noting that the trade-off between privacy and utility remains an unresolved constraint.
