AI News

Follow the latest AI and automation news

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

OpenAI now lets teams make custom bots that can do work on their own

OpenAI has announced the launch of cloud-based 'workspace' agents for users on Business, Enterprise, Edu, and Teachers plans. These agents can perform tasks like gathering product feedback from the web and sending reports via Slack, or drafting follow-up emails in Gmail. This initiative comes amid growing interest in AI agents, particularly after the viral success of OpenClaw, an AI agent known for its practical applications. Peter Steinberger, the founder of OpenClaw, has joined OpenAI, highlighting the company's commitment to enhancing its capabilities in this area.

Infrastructure

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Taiwan Semiconductor Manufacturing Company (TSMC) announced plans to open a packaging plant in Arizona by 2029.

Taiwan Semiconductor Manufacturing Company (TSMC) announced plans to open a packaging plant in Arizona by 2029. This comes as demand for advanced AI chips, like those made by Nvidia, continues to grow, highlighting the plant's significance in meeting market needs. TSMC is expanding its capabilities in Arizona, with construction already underway, as stated by Kevin Chang, the company's executive vice president. The new facility will feature advanced packaging technologies such as CoWoS and 3D integrated circuit manufacturing, reducing reliance on Taiwan.

Infrastructure

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Google Cloud launches two new AI chips to compete with Nvidia

Google has launched its new TPU processors, which are faster and cheaper than previous versions. This move aims to enhance cloud computing performance and reflects the company's commitment to providing advanced solutions in the sector. The new TPUs are part of Google's strategy to compete with Nvidia, which still holds a significant position in the cloud computing market. Despite the improvements offered by Google, the company continues to rely on Nvidia's technologies for some of its cloud services, highlighting the complexity of competition in this field.

💡

5 AI Models Tried to Scam Me. Some of Them Were Scary Good

What measures can be taken to enhance cybersecurity against AI threats?

Agents

Watch Sony’s elite ping-pong robot beat top-ranked players

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Agents

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Watch Sony’s elite ping-pong robot beat top-ranked players

Sony has unveiled Ace, the first robot capable of defeating top-ranked table tennis players while adhering to official International Table Tennis Federation rules. Unlike previous robots, Ace can compete against elite players and even win matches, showcasing significant advancements in sports robotics. Historically, robots like Omron's FORPHEUS have challenged amateur players, but Ace represents a breakthrough developed by Sony's AI division. It is engineered to match the speed and responsiveness of human players, setting a new standard in robotic sports competition.

Agents

Google turns Chrome into an AI co-worker for the workplace

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Agents

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Google turns Chrome into an AI co-worker for the workplace

Google has announced the launch of new auto browse features in Chrome, powered by Gemini, for enterprise users. This allows employees to automate tasks such as research and data entry, enhancing efficiency and saving time. The AI-driven technology analyzes data to provide accurate and rapid results, supporting organizations in making data-driven decisions. This launch comes at a time when the demand for tools that facilitate remote work and boost productivity is increasing.

Models

Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Models

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model

Qwen has launched its new model, Qwen3.6-27B, which outperforms its predecessor, Qwen3.5-397B-A17B, in all software performance metrics. The new model is more efficient, with a size of 55.6 GB compared to 807 GB for the previous version. Improvements in Qwen3.6-27B include superior performance in programming tasks, demonstrating advantages over all previous open models in tests. It has been successfully utilized with the Unsloth release, allowing users to easily leverage its advanced capabilities.

Governance

U.

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Governance

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

U.

U.S. Commerce Secretary Howard Lutnick announced that Nvidia's H200 chips have not yet been sold to Chinese companies due to difficulties in obtaining Chinese government approval. The central government is focusing investments on its domestic industry, preventing these purchases. In January, the Trump administration approved H200 sales under specific conditions, raising concerns in Washington about potential military applications by Beijing. However, shipments have stalled due to disagreements over sales terms between the U.S. and China.

Models

AutoAdapt: Automated domain adaptation for large language models

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Models

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

AutoAdapt: Automated domain adaptation for large language models

AutoAdapt is an automated framework designed to expedite the adaptation of large language models in specialized fields. This new system enables the automatic planning and selection of adaptation strategies, significantly reducing the time and costs associated with manual adaptation processes. AutoAdapt relies on an organized graph that outlines all adaptation steps, ensuring reliable and swift results. In sensitive areas such as law and medicine, the adaptation process requires quick and dependable responses, as errors can lead to severe consequences.

Business

Reversing enterprise security costs with AI vulnerability discovery

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Business

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Reversing enterprise security costs with AI vulnerability discovery

Mozilla Firefox's engineering team discovered 271 security vulnerabilities using Anthropic's Claude Mythos model. These vulnerabilities were addressed in version 150, following a previous collaboration with Anthropic that resulted in 22 fixes in version 148. This progress is transforming how organizations manage security threats. Integrating advanced AI models into continuous integration systems necessitates significant investments in infrastructure. Companies must establish secure database environments to manage the contextual windows required for large codebases, ensuring the protection of enterprise logic.