Vitalik Updates on Decentralized AI and CROPS AI Progress
Ethereum founder Vitalik Buterin shared the latest updates: DeepSeek V4 has been released, featuring a 2-bit quantized version that can run within 90GB of memory, but it operates slowly on AMD hardware (about 7 tok/s) and faster on Apple hardware (about 35 tok/s).
He emphasized that true support for multiple hardware vendors is a key distinction between "mere decentralized AI" and "genuine CROPS AI."
Additionally, the messaging-daemon has added Telegram support, Lucebox-Hub has significantly improved efficiency when running dense models, and the VoxTerm local AI recording tool is still under development.
Vitalik also pointed out that there are numerous cross-applications between the CROPS Ethereum access layer and CROPS AI, such as ZK-enabled paid calls to remote LLMs that can simultaneously address private RPC reading issues, and he called for the development of more Ethereum-specific fine-tuning models.
Source: Public Information
ABAB AI Insight
Vitalik has long been promoting the integration of Ethereum with decentralized computing and AI. This update continues his emphasis on CROPS AI (Compute Really Owned and Permissionlessly accessible by the People), focusing on hardware diversity, privacy computing, and application layer integration.
On the capital front, Vitalik encourages developers to invest resources in multi-hardware optimization, ZK privacy calls, and Ethereum-specific model fine-tuning, aiming to attract more developers to build a truly decentralized AI infrastructure by lowering hardware barriers and enhancing privacy capabilities, while also bringing new computing and payment demands to the Ethereum ecosystem.
With advancements in privacy Layer 2 solutions like Aztec Labs and the emergence of efficient small models like Mistral Leanstral, the Ethereum ecosystem is currently transitioning from a purely Layer 1 to a privacy + AI computing fusion layer.
Essentially, this represents a technological substitution and industrial chain reconstruction: CROPS AI seeks true people-owned and permissionless access computing, with mechanisms to support hardware diversity and ZK privacy layers to resolve centralization bottlenecks, accelerating the shift of capital and developers from cloud giant-dominated centralized AI to a distributed, privacy-protecting decentralized AI network, promoting the migration of computing resources from a few data centers to global personal devices and the Ethereum ecosystem.
ABAB News · Cognitive Law
Decentralization is not just a slogan, but truly supports multiple hardware, allowing ordinary people to run models. The worse the hardware compatibility, the more decentralization sounds like an empty phrase. A truly powerful AI future begins with fair access across hardware.