Scan to download
BTC $66,344.28 +1.24%
ETH $1,794.31 +4.60%
BNB $616.94 +0.49%
XRP $1.23 +4.13%
SOL $74.09 +4.12%
TRX $0.3187 -0.43%
DOGE $0.0879 -0.82%
ADA $0.1773 -2.79%
BCH $223.76 +6.50%
LINK $8.27 +1.70%
HYPE $67.41 +5.57%
AAVE $73.70 +7.85%
SUI $0.7920 +0.66%
XLM $0.2126 +13.13%
ZEC $513.22 +9.14%
BTC $66,344.28 +1.24%
ETH $1,794.31 +4.60%
BNB $616.94 +0.49%
XRP $1.23 +4.13%
SOL $74.09 +4.12%
TRX $0.3187 -0.43%
DOGE $0.0879 -0.82%
ADA $0.1773 -2.79%
BCH $223.76 +6.50%
LINK $8.27 +1.70%
HYPE $67.41 +5.57%
AAVE $73.70 +7.85%
SUI $0.7920 +0.66%
XLM $0.2126 +13.13%
ZEC $513.22 +9.14%

framework

DGrid AI released the latest research paper PoQ-Judge, completing the closed loop of decentralized LLM quality assessment with a multi-architecture evaluation framework

The decentralized AI infrastructure network DGrid AI today released its latest research paper "PoQ-Judge," proposing a multi-architecture quality assessment framework that does not require reference answers. This means that in real deployment environments, there are often no standard answers for comparison, yet the protocol can still reliably score the quality of model responses and allocate incentives accordingly. This is a key piece that has long been missing in DGrid's decentralized LLM inference quality assessment system.PoQ (Proof of Quality) is a consensus mechanism independently developed by DGrid, designed to prevent model providers from deploying low-quality models, fabricating data, or hiding computational costs at the protocol level, thereby ensuring service quality and pricing transparency. The DGrid team has been continuously working on PoQ and has published four research papers to date. The newly released PoQ-Judge has trained three assessment models covering different quality and cost scenarios, achieving a correlation of up to 0.747 with human scoring on the retention test set, significantly outperforming all previous reference answer-based evaluators, while reducing assessment costs by over 72% through cascading evaluation and online weight calibration.With the implementation of PoQ-Judge, the entire process from quality assessment → scoring → incentive allocation has completely eliminated reliance on reference answers, thus establishing a closed loop for the quality of decentralized LLM inference.DGrid AI is a decentralized AI intelligent network dedicated to building an open, transparent, and community-driven AI infrastructure. Focusing on model invocation and application experience, DGrid has launched several core products: the AI Gateway that aggregates mainstream large models globally, the one-click deployment platform for AI agents DClaw, the anonymous model competition platform AI Arena, and the intelligent model recommendation assistant Dori, providing one-stop services for developers and users. It is reported that DGrid AI's revenue has surpassed 20 million dollars in six months.

Hong Kong Securities and Futures Commission: Will continue to promote the construction of a regulatory framework for digital assets and support AI financial applications

According to Crowdfund Insider, the Chairperson of the Hong Kong Securities and Futures Commission (SFC), Laura Liang, stated at the Caixin Summer Summit that Hong Kong will continue to expand its digital asset regulatory framework and promote the application of artificial intelligence (AI) in the financial services sector to consolidate its position as an international financial center.Laura Liang pointed out that regulatory agencies will improve the institutional framework around areas such as digital asset trading, custody, investment consulting, and asset management, while adhering to the regulatory principle of "same business, same risks, same rules," achieving a balance between innovation and investor protection.She stated that as the application of AI in the financial industry accelerates, regulatory focus will include potential risks such as model reliability, algorithm bias, data privacy, and cybersecurity, emphasizing that financial institutions need to strengthen risk management during the innovation process.In addition, the Hong Kong Securities and Futures Commission and relevant regulatory agencies have expanded the regulatory sandbox mechanism, allowing financial institutions to test generative AI applications in a controlled environment to promote technological implementation and compliant development. Analysts believe that Hong Kong is further enhancing the openness and standardization of its financial markets through a dual regulatory framework for digital assets and AI, while also increasing its competitiveness in the global capital markets.

The robot AI data platform Mecka AI has completed a $60 million financing round, led by Framework Ventures

According to Fortune, the startup Mecka AI, which focuses on training data for robotic AI, announced the completion of a total financing of $60,000,000, including a $25,000,000 Series A round completed last November and a subsequent $35,000,000 follow-on financing. Both rounds were led by Framework Ventures, with participation from institutions such as Menlo Ventures, SV Angel, and Kindred Ventures.Mecka AI primarily collects human motion data through body sensors, iPhones, and custom hardware, including physical behavior data such as gestures and gait, to train robotic AI models.The company's founder and CEO, Josh Gao, stated that its core philosophy is to train robots using human behavior data rather than traditional teleoperation data, thereby enhancing the general capabilities of robots in the real world.It was introduced that Mecka AI was established in 2025 and currently has about 40 employees. The company claims that based on signed contracts, its annual recurring revenue (ARR Run Rate) is expected to reach $100,000,000, but it has not disclosed a specific client list.Vance Spencer, co-founder of Framework Ventures, stated that Mecka AI is one of the fastest-growing companies in the institution's portfolio.In the future, Mecka AI plans not only to provide training data but also to directly participate in the training and deployment of robotic models, promoting the commercialization of robots in real-world scenarios.

first_img Aave Labs has released an ARFC proposal aimed at establishing a unified standardized framework for the listing of technical assets

Aave Labs has released an ARFC proposal, suggesting the establishment of a standardized technical asset listing framework for Aave V3, V4, and Aave Horizon, setting unified technical requirements for asset listing, parameter expansion, and ongoing monitoring. The framework covers core areas such as ERC20 compatibility, oracles, permission control, minting and burning logic, pause and blacklist mechanisms, upgradability, exchange rates and yield mechanisms, token architecture, cross-chain bridge risks, audit and security history, and external dependencies. This framework does not replace market risk analysis and governance judgment but provides a technical qualification baseline.The framework aims to address "hidden risks" such as unlimited issuance, weak upgrade permissions, inconsistent bridging supply, opaque redemption paths, and reliance on off-chain custody. These issues may directly threaten the protocol's solvency, liquidation systems, and collateral parameter security. The framework particularly emphasizes additional scrutiny for cross-chain assets, yield-bearing assets, and off-chain dependent assets such as RWAs, including bridge structures, off-chain legal arrangements, custody mechanisms, and supply integrity. Assets with significant technical flaws may face reduced borrowing limits, restricted collateral parameters, delayed launches, or even recommendations to deny access to the protocol in the future.
app_icon
ChainCatcher Building the Web3 world with innovations.