Judge denies Musk’s attempt to block OpenAI from becoming a for-profit entity

A federal judge in California denied Elon Musk’s attempt to block OpenAI from converting into a for-profit entity, ruling that Musk and his legal team failed to meet the burden of proof for such an extraordinary injunction. However, other aspects of Musk’s lawsuit against OpenAI, which he co-founded with Sam Altman in 2015, will still proceed. The dispute stems from OpenAI’s transition from a nonprofit AI research lab to a more commercially driven entity, a shift that Musk has vehemently opposed.

Musk initially sued OpenAI in March 2024, claiming the company breached its fiduciary duty and contract agreements. The lawsuit was later expanded to include allegations that OpenAI and Microsoft violated antitrust laws by preventing investors from funding rival AI firms like Musk’s startup, xAI. In December 2024, OpenAI defended its restructuring, arguing that its move toward a public benefit corporation in 2025 was necessary to attract the massive investments required for AI development.

In February 2025, Musk and a group of investors offered to buy OpenAI’s nonprofit parent for $97.4 billion, but Altman rejected the bid, calling it an attempt to slow down competition. Musk responded by stating he would withdraw his acquisition offer if OpenAI abandoned its for-profit transition. In the ongoing legal battle, OpenAI has pointed out that Musk himself advocated for a for-profit model in 2017, contradicting his current stance. Despite the court’s decision against blocking the transition, Musk’s broader claims against OpenAI will continue through federal court.

Get More From Every Word

Meet VoiceHub from Rev - the #1 platform for recording, transcribing, and analyzing speech. It’s revolutionizing how businesses handle their most valuable asset: conversations. 

What sets VoiceHub apart? Their AI accuracy beats Microsoft, Google, and other leading providers in enterprise environments.

But it's not just about accuracy—it's about changing the way you use those accurate transcripts. With VoiceHub, you get:

• Universal capture of audio and video across mobile, desktop, and meetings

• Best-in-class AI transcription and summaries in seconds

• Enterprise-grade security with HIPAA, SOC 2 Type II compliance, and SSO

• Seamless integration with 18 major tools like Zoom and Slack

NextGenAI: A consortium to advance research and education with AI

OpenAI has launched NextGenAI, a groundbreaking consortium bringing together 15 leading research institutions to advance AI-driven research and education. Backed by $50 million in research grants, compute funding, and API access, this initiative aims to accelerate breakthroughs across various fields, from healthcare and manufacturing to education and digital libraries. Institutions such as Harvard, MIT, Oxford, and The Ohio State University will use AI to tackle high-impact challenges, including diagnosing rare diseases, developing new AI-powered learning models, and digitizing historical texts. By fostering collaboration between academia and industry, OpenAI seeks to push the boundaries of AI innovation and ensure that researchers and students have the tools they need to shape the future of AI.

A key focus of NextGenAI is empowering students and educators to become AI-fluent. Texas A&M, MIT, and Howard University will use OpenAI’s resources to integrate AI into curricula, enhance learning experiences, and provide hands-on training for students. Meanwhile, institutions like the University of Mississippi and the Boston Public Library are exploring ways to use AI to improve education and accessibility. This initiative follows OpenAI’s ChatGPT Edu launch in 2024, which provided universities with access to AI-powered learning tools. Now, with NextGenAI, OpenAI is further cementing its role in AI education by equipping institutions with advanced tools and funding to drive critical innovation.

While NextGenAI is positioned as a boost for academia, it also strengthens OpenAI’s foothold in the AI research ecosystem. The program arrives at a time when U.S. AI research funding faces uncertainty, with recent policy shifts affecting agencies like the National Science Foundation. By providing AI resources and funding, OpenAI is ensuring that universities continue to make progress despite these challenges. However, the initiative also serves OpenAI’s strategic interests, encouraging researchers to adopt its proprietary AI tools over open-source alternatives. As NextGenAI unfolds, its impact on both academic research and the broader AI landscape will become clearer, shaping the next generation of AI-driven discoveries.

People are using Super Mario to benchmark AI now

Researchers at the University of California San Diego’s Hao AI Lab recently tested AI models by making them play Super Mario Bros. using a custom-built framework called GamingAgent. The experiment, which ran the game in an emulator, required AI models to generate Python code to control Mario based on in-game screenshots and basic instructions like dodging obstacles. Among the tested models, Anthropic’s Claude 3.7 performed best, followed by Claude 3.5, while OpenAI’s GPT-4o and Google’s Gemini 1.5 Pro struggled. Interestingly, reasoning-based models, which typically excel in benchmarks, performed worse due to their slower decision-making speed, which proved disadvantageous in a fast-paced game like Super Mario Bros.

While AI has long been tested using video games, this study highlights the challenges of real-time decision-making. Unlike structured AI tasks, Super Mario Bros. requires split-second reflexes and strategic planning, forcing AI to “learn” gameplay rather than simply recognizing patterns. The research suggests that AI models with rapid response times, even without deep reasoning abilities, may have an advantage in dynamic environments. However, it also raises questions about the validity of gaming as a benchmark for AI progress, as games provide controlled, abstract scenarios with unlimited training data—far from the complexities of real-world applications.

This skepticism echoes concerns voiced by AI experts like OpenAI’s Andrej Karpathy, who recently called the current state of AI evaluation an “evaluation crisis.” With AI models demonstrating unpredictable strengths and weaknesses across different benchmarks, researchers are struggling to determine how best to measure true AI capability. While gaming benchmarks are engaging and provide useful insights, they may not fully capture AI’s real-world potential. Still, at the very least, they offer an entertaining way to watch AI in action.

Other stuff

• ChatGPT is made from 100 million of these [The Perceptron]

• Google still limits how Gemini answers political questions

• Klarna CEO doubts that other companies will replace Salesforce with AI

• Amazon is reportedly developing its own AI ‘reasoning’ model

• Call centers using AI to ‘whiten’ Indian accents

• A New AI Worry: Many Young Coders No Longer Know How Their Code Really Works

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.

OR