NerdNews

OpenAI has released GPT-5.2, its newest family of AI models, in response to Google's Gemini 3 model. GPT-5.2 features a 400,000-token context window and a knowledge cutoff date of August 31, 2025. The model is available in three versions: Instant, Thinking, and Pro, and is rolling out to paid ChatGPT subscribers and developers via API.

Every's compound engineering process involves a 4-step loop: plan, work, review, and compound. Agents write code and create tests according to plans, and the engineer reviews the output and feeds the results back into the system to improve future development.

Disney has invested $1 billion in OpenAI and licensed 200 characters for use in OpenAI's Sora AI video generator. The deal allows users to create short videos featuring Disney, Marvel, Pixar, and Star Wars characters, and also includes the use of ChatGPT for Disney employees and the deployment of OpenAI technology to build new features for Disney+.

Google released a 'reimagined' version of its research agent Gemini Deep Research based on its state-of-the-art foundation model, Gemini 3 Pro. The new agent allows developers to embed Google's research capabilities into their own apps and has been tested on various benchmarks, including DeepSearchQA and Humanity's Last Exam. OpenAI also launched its GPT 5.2 model on the same day, which it claims bests Google's model on several benchmarks.

Google's AI try-on feature now allows users to virtually try on clothes using just a selfie. The feature uses Nano Banana, Google's Gemini 2.5 Flash Image model, to generate a full-body digital version of the user for virtual try-ons.

Rivian announced a new AI assistant, lidar technology, and autonomous driving features coming in 2026. The company is developing its own custom processor and autonomy platform, which will enable hands-free driving on over 3.5 million miles of US roads. The new technology will be available on the upcoming R2 midsize electric SUV, with a starting price of $45,000.

China has launched a 34,175-mile AI network, called the Future Network Test Facility (FNTF), which connects 40 cities and allows the country's top computing centers to operate as a unified system. The network achieves 98% of a single center's efficiency and can support 128 diverse networks and 4,096 service trials. It has the potential to significantly boost AI development in China.

The article discusses the potential effects of AI investment on urban digital twins. With the current AI investment bubble, there are concerns that it may burst, affecting the development of digital twins. However, analysts believe that AI will continue to play a major role in transforming industries, including the development of digital twins. Digital twins can benefit from AI in various ways, such as making data usable and creating data for unmeasurable phenomena. The article also highlights the potential of digital twins in urban planning, with cities like Osaka, Japan, developing smart cities that integrate AI, extended reality, and decentralized physical infrastructure networks.

Runway's GWM-1 is a significant step in building 'world models' that can simulate environments and remain coherent for minutes. The models have potential applications in game development, virtual reality, and robotics. Runway is entering a competitive field with GWM-1, but its claims of consistency and coherence are impressive if true.

OpenID Connect (OIDC) simplifies user authentication but introduces new security risks. Common pitfalls include treating Client Secrets as configuration, incomplete JWT validation, and misconfigured Trust Policies in Non-Human Identity (NHI) workflows. Developers must prioritize secrets management, token validation, and operational vigilance to secure OIDC integrations.

Oracle has posted a 13% year-over-year revenue increase, with significant growth in cloud infrastructure and applications. The company expects OCI margins to increase by 30-40% over the life of a customer contract, driven by accelerated capacity delivery and improved profitability for AI workloads.

Rivian unveiled its plans for autonomous driving, including custom silicon, lidar, and a potential entry into the self-driving ride-hail market. The company will expand its hands-free driver-assistance software and plans to allow drivers to take their eyes off the road. Rivian's goal is to achieve 'personal L4' autonomy, enabling cars to operate without human intervention in specific areas.

Runway has released its first world model, GWM-1, which works through frame-by-frame prediction to create simulations with an understanding of physics. The company has also updated its Gen 4.5 model with native audio and long-form, multi-shot generation capabilities. Additionally, Runway has introduced GWM-Worlds, GWM-Robotics, and GWM-Avatars, which are separate models that can be used for various applications such as gaming, robotics, and human behavior simulation.

Spoor's bird-monitoring AI software uses computer vision to track and identify bird populations and migration patterns, helping wind farms reduce their impact on local bird populations. The software can detect birds within a 2.5-kilometer radius and has improved its bird identification accuracy to about 96%. Spoor now works across three continents and with more than 20 of the world's largest energy companies.

Harness, an AI DevOps tool, has raised $240 million in a Series E funding round, valuing the company at $5.5 billion. The startup aims to automate the 'after-code' gap in software development, using AI agents to automate functions like testing, verification, security, and governance. With over 1,000 enterprise customers, Harness plans to expand its R&D efforts and hire hundreds of engineers to build out additional automated testing, deployment, and security capabilities.

A security analysis of Docker Hub images has found over 10,000 public container images exposing sensitive secrets, including live cloud keys and credentials, from over 100 companies. The exposed secrets grant access to production systems, cloud services, CI/CD pipelines, and AI platforms. Developers are urged to use dedicated secrets management tools and vaults to prevent such leaks.

The federal government is considering a moratorium on state-level regulation of AI, citing the need for a unified national approach to avoid a 'patchwork' of regulations. However, this argument is flawed, as other industries have successfully navigated local regulations. States should be allowed to regulate AI to protect their citizens' interests and drive innovation. A federal ban would concentrate power in the hands of a few large AI companies and undermine democracy.

Dozens of state and territorial attorneys general have sent a warning letter to Big Tech companies, including OpenAI, Microsoft, and Anthropic, to mitigate harm caused by 'sycophantic and delusional' AI outputs, particularly those that affect children. The letter lists disturbing behaviors, such as AI bots pursuing romantic relationships with children, and suggests remedies like developing policies to mitigate dark patterns in AI products.

Port, an Israeli startup, has raised $100M at an $800M valuation to take on Spotify's Backstage with its proprietary dev tool portal. The platform helps companies manage AI agents, providing a layer of orchestration with features like measuring agent performance and adding human approval processes. With big-name customers and tier-one VCs, Port is a startup to watch in the agentic management space.

President Trump signed an executive order that creates a national regulatory framework for artificial intelligence, while also challenging state AI laws. The order establishes an AI litigation task force within the Justice Department and directs the Department of Commerce to craft guidelines that could make states ineligible for future broadband funding if they pass 'onerous' AI laws.