AI News

AI News Brief: South Korea’s tech sector is posting record profits, but new guidelines could make it harder for workers to share in the gains. Samsung Electronics union members protested pay levels outside the Pyeongtaek semiconductor plant on April 23, underscoring the tension. Under the new guidelines, companies have no obligation to negotiate profit-sharing with workers, effectively weakening unions’ bargaining leverage over AI chip-driven windfalls at the regulatory level. See the original article for full details.

AI news brief: OpenAI’s latest case study shows game developer Playco using the GPT-6 Astra model for game prototyping, starting from a single ‘grey box’ (the most basic playable skeleton) and branching it out into three prototypes with distinct themes. Compared to their workflow with the previous-generation model, Playco’s team reported a 50% drop in manual fixes needed after switching to GPT-6 Astra.

AI news brief: Lithuanian legal tech company Legora integrated OpenAI’s latest GPT-6 Astra model into its financial statement review workflow, and testing showed the system completed a 41-document review in minutes, catching all four intentionally planted errors, with review performance improving nearly 40% over the previous setup.

AI news flash: OpenAI announced ‘Daybreak for Frontline Defenders,’ a $1 billion commitment to expand frontline defense units’ access to cutting-edge cybersecurity AI, training resources, and support services, targeting teams that keep essential services running so they can strengthen their defenses against cyber threats.

AI News Brief: Nvidia has announced it’s acquiring open-source AI platform Hugging Face for $12.93 billion, officially expanding into AI software and developer tools. Hugging Face currently serves over 18 million developers and hosts more than 3 million AI models on its platform. Nvidia CEO Jensen Huang…

AI news brief: GitHub Blog published a beginner-focused tutorial on running multiple AI agents (parallel agents) at the same time in the GitHub Copilot app, aiming to help newcomers get past the initial discomfort of running several agents simultaneously and discover the productivity gains of parallel workflows.

AI News Brief: this is Shimizuwan’s daily translation task, outputting a Traditional Chinese summary directly. Google DeepMind and Google Research today launched WeatherNext 3, billed as the most advanced and accurate global weather AI model out there, ranked best in the industry per real-time evaluation results from independent group Brightband. The model’s biggest breakthrough is that it learns directly…

AI News Brief: This news summary is limited in detail, only touching on the background of the internet trust crisis without providing specific mechanisms or numbers — the internet is facing a trust crisis, and not just because social media is flooded with low-quality AI-generated content (AI slop). More seriously, AI-generated text and images have infiltrated job applications, product reviews, and even insurance claims, making it hard for both platforms and users to tell what’s real.

AI news brief: this one is short and lacks concrete mechanism details (it only touches on the theme that AI detection is harder than telling real from fake), so I’m handling it per rules 2/3. AI-generated content is flooding job applications, product reviews, and even insurance claims, making it harder for both platforms and users to tell what’s real, and the online trust problem now goes well beyond social media getting swamped by AI slop. Over the past few years, several startups have jumped into AI detection, trying to tackle this increasingly serious identification challenge.

AI news brief: OpenAI’s new model ‘Astra,’ according to a Tuesday report from The Information, will use a reasoning technique called ‘recurrent depth’ (also known as ‘opaque recurrence’) that breaks from the linear, step-by-step thinking most reasoning models rely on. The technique lets the model process the same query in a loop multiple times instead of unfolding steps sequentially, leaving fewer readable traces and effectively bypassing traditional chain-of-thought logging - which could make the model’s reasoning harder to monitor.