If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.

This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.

Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.

Microsoft and OpenAI announced an amended agreement that gives both companies more flexibility, while keeping Microsoft deeply tied to OpenAI’s future.

The key change: Microsoft remains OpenAI’s primary cloud partner, but no longer has the same exclusive grip. OpenAI products will still launch first on Azure unless Microsoft cannot or chooses not to support the needed capabilities, but OpenAI can now serve all of its products to customers across any cloud provider. That is a major shift because it opens the door for OpenAI to become much more cloud-flexible while still keeping Azure as the default first stop.

Microsoft also keeps a license to OpenAI’s model and product IP through 2032, but that license is now non-exclusive. Microsoft will no longer pay revenue share to OpenAI, while OpenAI will continue revenue-share payments to Microsoft through 2030, at the same percentage but with a total cap. Microsoft also remains a major shareholder, meaning it still benefits if OpenAI grows into an even bigger company.

Right after the Microsoft-OpenAI partnership update, AWS announced a major OpenAI expansion of its own.

Amazon says the latest OpenAI models are coming to Amazon Bedrock in limited preview, meaning AWS customers can access OpenAI models through the same Bedrock APIs, governance tools, and security controls they already use. This puts OpenAI models beside Anthropic, Meta, Mistral, Cohere, Amazon’s own models, and others inside one enterprise AI platform.

The most interesting part is not just model access. AWS is also bringing Codex on Amazon Bedrock in limited preview. Amazon says more than 4 million people use Codex weekly for coding tasks like refactoring, explaining systems, writing tests, and automating development work. Now enterprise teams can use Codex through AWS credentials, Bedrock infrastructure, the Bedrock API, Codex CLI, Codex desktop app, and VS Code extension.

AWS is also launching Amazon Bedrock Managed Agents, powered by OpenAI, also in limited preview. This is aimed at companies that want production-ready AI agents without stitching everything together themselves. These agents include infrastructure pieces like persistent memory, skills, identity, permissions, audit logs, and compute options. Amazon says each agent has its own identity, logs actions for auditability, and runs inference on Amazon Bedrock.

Google has reportedly signed a classified agreement with the U.S. Department of Defense that allows the Pentagon to use Google AI models for “any lawful government purpose.”.

The agreement reportedly places Google alongside other AI companies that have made similar classified AI deals with the U.S. government, including OpenAI and xAI. Classified networks can involve sensitive work such as mission planning and weapons targeting, which explains why this deal is drawing so much attention.

The reported contract includes language saying the AI system is not intended for domestic mass surveillance or autonomous weapons without appropriate human oversight and control. But there is a major caveat: the agreement reportedly does not give Google the right to control or veto lawful government operational decisions. Google is also reportedly required to help adjust AI safety settings and filters at the government’s request.

This is already causing internal backlash. More than 600 Google employees reportedly signed an open letter asking CEO Sundar Pichai not to make Google’s AI systems available for classified workloads, warning that the technology could be used in harmful ways. This echoes Google’s older Project Maven controversy, when employee protests pushed the company to step away from a Pentagon AI contract in 2018.

World-R1: Reinforcing 3D Constraints for Text-to-Video Generation

Smart computer models use extra power to think through problems step by step. Shortening this process often causes mistakes. This paper introduces a clever filtering tool removing useless thoughts while keeping helpful ones. This allows the model to give faster answers without sacrificing accuracy or stability during its training process.

Stabilizing Efficient Reasoning with Step-Level Advantage Selection

Programs that create videos from simple typed text often produce warped or physically impossible shapes. This paper introduces a teaching method using rewards to fix this. By grading the program on realistic geometry, the method enforces strict physical rules without altering original code, resulting in accurate and fluid video scenes.

ProEval: Proactive Failure Discovery and Efficient Performance Estimation for Generative AI Evaluation

Testing artificial intelligence programs is often slow and extremely expensive. This paper presents a proactive evaluation system that uses past knowledge and mathematical predictions to guess how well a program works. This smart method actively finds severe mistakes using significantly less data, efficiently judging the overall safety and system logic.

Nvidia's New AI Agent System Nvidia released a new artificial intelligence system that can see, hear, and read all at the exact same time. By combining these digital senses into a single program instead of using separate ones, automated computer helpers can understand videos, audio, and text much faster. 

Anthropic's Claude for Creatives Anthropic connected its artificial intelligence assistant directly into popular creative software programs used for making art, music, and 3D designs. Artists can now type simple instructions to have the computer handle boring chores, write custom digital tools, or smoothly move files between different applications. 

Mistral's Reliable Business Workflows Mistral created a new tool to help companies safely use artificial intelligence for important daily operations without the computer programs crashing or making invisible mistakes. It acts as a strict digital manager that tracks every step of a complicated task and can automatically pause to ask a human for approval before continuing. 

OpenAI's Symphony System OpenAI engineers built a system that lets artificial intelligence programs act like independent workers that pull their own assignments from a company's shared to-do list. Instead of humans constantly supervising the computer, the system automatically hands out coding jobs, checks for mistakes, and keeps the digital workers running day and night. This radically sped up the team's progress because human programmers only had to review the final product instead of typing the code themselves.

Meta's Space Solar and Energy Storage Meta is funding bold new power technologies to secure enough constant, clean electricity to run its massive computer centers. The company is paying to develop satellites that collect solar power in space and beam it down to Earth, allowing existing solar farms to produce electricity even when it is dark outside. They are also investing in advanced batteries that can store this green energy for many days instead of just a few hours.