OpenAI releases ChatGPT Images 2.0. Text generation is more accurate, and posters and portrait designs look more beautiful. A new “thinking mode” enables real-time web browsing, batch output, and self-review. With a single prompt, you can generate up to eight images; it can maintain character and style consistency. The maximum resolution is 2K, and the aspect ratio can be selected from 3:1 to 1:3. Asian-language optimization is significantly improved. It offers an API and paid commercial plans. Knowledge cutoff is December 2025, and even complex compositions still require a few minutes.

Ripple CEO Brad Garlinghouse publicly praised the regulatory shift driven by the newly appointed Securities and Exchange Commission (SEC) chair, Paul Atkins, on April 20, describing it as “a refreshing breath of sanity that helps people regain their composure.” Garlinghouse viewed the SEC’s policy pivot as the key driver behind an improvement in sentiment across the U.S. crypto market.

Curve Finance has been attacked over LayerZero infrastructure related to rsETH, and has temporarily suspended cross-chain functionality to prevent risk, impacting CRV cross-chain bridging and the fast bridging of crvUSD. Founder Egorov said the incident demonstrates the risk of “non-isolated lending,” and proposed a fully isolated mode as an alternative. Kelp DAO also suffered losses of about $292 million due to the attack, affecting lending activity on the Aave platform.

A study by the University of California reveals security vulnerabilities in the supply chain of large language models (LLMs), especially malicious man-in-the-middle attacks that third-party routers may carry out. The research found that 26 routers injected malicious commands to steal credentials and sensitive data. Users have difficulty noticing the boundary between credential handling and theft, and the “YOLO mode” further increases security risk. The study recommends that developers isolate sensitive operations and choose router services with transparent auditing to strengthen protection.

Google announced on April 10 that its AI Mode restaurant reservation feature has expanded to eight markets, including Australia, Canada, Hong Kong, India, New Zealand, Singapore, South Africa, and the United Kingdom, signaling that the adoption of agentic AI is being further deepened. This feature allows users to search for restaurants, check availability, and book with a single click using natural language, improving the convenience of online reservations. The addition of Asian markets highlights Google’s presence in the region, but Taiwan has not been included in the initial rollout list.

BASIS announced that its private testing phase has been successfully completed, demonstrating execution latency of under 50 microseconds and 100% operational reliability. The platform uses a controlled rollout strategy, focusing on performance validation, and will set a new benchmark for institutional participants. In the future, it will continue to maintain a “highly filtered” mode to ensure outstanding system performance.

DeepSeek quietly pushed a three-mode test on April 8: fast, expert, and vision—three-track routing. The community sees it as the final warm-up before a formal V4 launch. (Background recap: DeepSeek V4 refuses NVIDIA and goes looking for Huawei! Alibaba, ByteDance, and Tencent are all rushing to buy the Ascend 950PR chips) (Additional context: DeepSeek V4 announces it will give up NVIDIA! How far has China’s AI “compute independence” breakthrough campaign gotten?) Table of contents Toggle Three-track routing: fast, expert, and vision, each doing its own job Controversy in expert mode: is it an architecture difference, or prompt engineering? The real meaning of V4: if compute power is truly decoupled At 1:00 a.m. on April 8, the DeepSeek website and

In brief A developer discovered that forcing Claude to speak like a caveman slashes output tokens, and therefore costs, by up to 75%. The internet immediately turned it into a GitHub skill. With Anthropic charging so high per output tokens, grunt-mode is less of a joke and more of a

High-risk altcoins often outperform during late-stage bull cycles when liquidity expands rapidly. Infrastructure tokens like Chainlink gain relevance as decentralized applications. The risks are also high, but their location denotes a high-yield, dynamic part of the market that would almost a

Magic Eden is discontinuing its wallet app, transitioning to an export-only mode, which means users must export their private keys to avoid losing access to their funds. This change emphasizes the importance of self-custody and key preservation for cryptocurrency users.