The U.S. Third Circuit Court of Appeals issued a landmark ruling against ROSS Intelligence, finding that the company infringed Thomson Reuters's copyright by using Westlaw headnotes to train its AI legal research tool. The September 29, 2026 decision affirmed a lower court ruling and determined that ROSS's use of the material was not sufficiently transformative to qualify as fair use. This marks the first federal appellate decision directly addressing copyright fair use in the AI training context. ROSS shut down in 2021 after the lawsuit was filed due to financial strain from the litigation.
NVIDIA launched AICR v1.0, an open and verifiable system for configuring GPU-accelerated Kubernetes clusters. The tool provides four core capabilitiesāSnapshot, Recipe, Bundle, and Validationāto help teams manage compatible versions across components like host kernels, GPU drivers, and container runtimes. The release includes committed compatibility baselines, live-cluster validation, signed evidence, and supply-chain verification, covering major Kubernetes services and NVIDIA's current accelerator portfolio.
Mark Zuckerberg acknowledged that Llama 4 failed because Meta used a large team model similar to Instagram rather than keeping tight-knit groups. He explained that AI models require smaller, focused teams to succeed. Meta subsequently created the Meta Superintelligence Lab with Alexandr Wang to rebuild its AI efforts going forward.
Flagship Pioneering introduced iris labs, a company building personal AI products focused on human wellbeing. Its first product, iris chat, is an AI thought partner available on iMessage and RCS that uses expert-informed guidance combined with AI judges to handle emotionally complex conversations. A parallel AI safety agent assesses interactions in real time as a safety scaffold.
Abu Dhabi's Artificial Intelligence and Advanced Technology Council met to review AI progress, chaired by Sheikh Tahnoon bin Zayed. The council discussed the UAE-US AI Campus and Abu Dhabi's goal to become the world's first fully AI-native administration by 2027. The emirate aims to develop underlying AI technology and shape its economy around it.
Delhi University launched a three-month AI Film Making certificate course through its Campus of Open Learning. The course costs Rs 15,000 and offers 30 seats, with registration running from October 5 to October 31, 2026. Students will learn storyboarding, prompt writing, video editing, and audio production using AI tools.
Jon Stewart criticized Donald Trump and tech leaders over their stance on AI, questioning whether letting tech giants police themselves is viable. He mocked claims that AI will create universal high income and noted that tech leaders struggle to answer questions about jobs and AI.
Researchers developed a new framework to study how large language models react to repeated context, finding that internal predictions do not drift endlessly when given the same information repeatedly. Instead, models reach a stable state that determines whether context can change their final answer. Sometimes repeated statements fail to alter predictions, while other times a change becomes unavoidable.
A senior enterprise executive warned that organizations are solving for the wrong thing when it comes to AI adoption. The article identifies four types of gaps behind poorly supported decisions: Capability Gap, Design Gap, Delivery Gap, and Connection Gap. Only the Capability Gap is fixed by building more AI capability, while the others require different solutions.
Fast food businesses in London and New York are using older free AI models to generate food images for promotions, but the results often look unrealistic with strange patterns and textures. AI expert Jon Krohn said humans are sensitive to food that does not look real. Businesses turn to AI because it is cheaper than hiring photographers, but customers are increasingly noticing the difference.
Key Takeaways
- The Third Circuit ruled that ROSS Intelligence infringed Thomson Reuters's copyright by using Westlaw headnotes to train AI, marking the first federal appellate decision on AI training fair use.
- NVIDIA released AICR v1.0, an open system for configuring GPU-accelerated Kubernetes clusters with Snapshot, Recipe, Bundle, and Validation capabilities.
- Mark Zuckerberg said Llama 4 failed due to Meta's large team structure and the company created the Meta Superintelligence Lab with Alexandr Wang to rebuild AI efforts.
- Flagship Pioneering launched iris labs, offering iris chat, an AI thought partner on iMessage and RCS with real-time safety assessment.
- Abu Dhabi aims to become the world's first fully AI-native administration by 2027, as reviewed by its AI and Advanced Technology Council.
- Delhi University introduced a three-month AI Film Making course costing Rs 15,000 with 30 seats, covering storyboarding, prompt writing, and video editing.
- Jon Stewart criticized tech leaders and Trump over AI, questioning self-regulation and challenging claims about AI creating universal high income.
- A new study found that large language models reach a stable state when given repeated context, rather than drifting endlessly in their internal predictions.
- Organizations should diagnose Capability, Design, Delivery, and Connection gaps before buying more AI tools, as only the Capability Gap is fixed by adding AI capability.
- Fast food businesses in London and New York use older free AI models for food images, but customers notice unrealistic results despite lower costs compared to photography.
Third Circuit Rules Against ROSS Intelligence in AI Copyright Case
The U.S. Third Circuit Court of Appeals ruled that ROSS Intelligence infringed Thomson Reuters's copyright by using Westlaw headnotes to train its AI legal research tool. The September 29, 2026 decision affirmed a lower court ruling in the case Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc. The court found that ROSS's use of the headnotes was not sufficiently transformative to qualify as fair use. This marks the first federal appellate decision directly addressing copyright fair use in the AI training context.
Court Rules AI Training Using Copyrighted Material Is Not Fair Use
A U.S. appeals court found that Ross Intelligence's use of Thomson Reuters's copyrighted headnotes to train its AI model did not qualify as fair use. Ross shut down in 2021 after the lawsuit was filed due to financial strain from the litigation. The Third Circuit court of appeals issued a first-of-its-kind ruling with significant implications for the AI industry. The case highlights ongoing legal battles between rightsholders and tech companies over the use of intellectual property to train AI models.
NVIDIA Releases AICR v1.0 for GPU Cluster Configuration
NVIDIA launched AICR v1.0, an open and verifiable system for configuring GPU-accelerated Kubernetes clusters. AICR provides four core capabilities: Snapshot, Recipe, Bundle, and Validation to help teams manage compatible versions across components like host kernels, GPU drivers, and container runtimes. The v1.0 release includes committed compatibility baselines, live-cluster validation, signed evidence, and supply-chain verification. It covers major Kubernetes services and the current NVIDIA accelerator portfolio.
Flagship Pioneering Launches iris labs AI Wellbeing Company
Flagship Pioneering announced the launch of iris labs, a company building personal AI products to advance human wellbeing. The company's first product is iris chat, an AI thought partner designed to help users lead more fulfilling lives through conversation on iMessage and RCS. iris chat uses expert-informed guidance combined with AI judges to handle emotionally complex conversations. The product also includes a safety scaffold with a parallel AI safety agent that assesses interactions in real time.
Enterprises Should Diagnose Gaps Before Buying More AI Tools
A senior enterprise executive highlighted that existing AI models often have no connection to the decisions they are meant to support. The article identifies four types of gaps behind poorly supported decisions: Capability Gap, Design Gap, Delivery Gap, and Connection Gap. Only the Capability Gap is fixed by building more AI capability, while the others require different solutions. The author argues that organizations are consistently solving for the wrong thing by defaulting to more AI tooling.
Zuckerberg says Llama 4 failed due to team structure
Mark Zuckerberg said Llama 4 failed because Meta used a large team model like Instagram instead of a small group. He explained that AI models need tight-knit teams. Meta created the Meta Superintelligence Lab with Alexandr Wang to rebuild its AI efforts.
Jon Stewart tells Trump and AI leaders to stop lying
Jon Stewart criticized Donald Trump and tech leaders for their stance on AI. He questioned letting tech giants police themselves. Stewart mocked claims that AI will create universal high income. He also noted that tech leaders struggle to answer questions about jobs and AI.
Delhi University launches AI filmmaking certificate course
Delhi University introduced a three-month AI Film Making course through its Campus of Open Learning. The course costs Rs 15,000 and has 30 seats. Registration started on October 5, 2026, and ends on October 31, 2026. Students will learn storyboarding, prompt writing, video editing, and audio production using AI tools.
Abu Dhabi reviews AI progress and future goals
Abu Dhabi's Artificial Intelligence and Advanced Technology Council met to review AI progress. Sheikh Tahnoon bin Zayed chaired the meeting. The council discussed the UAE-US AI Campus and the goal to become the world's first fully AI-native administration by 2027. Abu Dhabi aims to develop underlying AI technology and shape its economy around it.
Why AI generated food images look so wrong
Fast food businesses in London and New York are using older free AI models to generate food images for promotions. These images often look unrealistic with weird patterns and textures. AI expert Jon Krohn said humans are sensitive to food that does not look real. Businesses use AI because it is cheaper than hiring photographers, but customers are noticing the difference.
Study reveals limits of context in AI models
Researchers created a new framework to study how large language models react to repeated context. They found that internal predictions do not drift endlessly when given the same information over and over. Instead, the models reach a stable state that decides if the context can change their final answer. Sometimes repeated statements fail to alter predictions, while other times a change becomes unavoidable. The study confirms these findings through experiments and offers new ways to understand AI behavior.
Sources
- Training Day: Third Circuit Affirms Westlaw Headnotes Decision in AI Copyright Fair Use Case
- Court Finds That Using Copyrighted Material as AI Training Data Is Not Fair Use
- AICR v1.0: Open, stable, and verifiable GPU cluster configuration
- Flagship Pioneering Launches iris labs to Transform Human Wellbeing Through AI
- Before you buy more AI, diagnose the gap you actually have
- Zuckerberg Blames Team Structure For Llama 4ās Failure, Says AI Models Needs Small, Tight-Knit Teams
- Jon Stewart on Trump and AI tech bros: āStop lying to usā
- Delhi University Launches AI Film Making Certificate Course, Check Duration And Skills
- Abu Dhabi's Artificial Intelligence and Advanced Technology Council meets to review AI progress
- Why A.I. Food Slop Looks So Gross
- Not All Answers Are Contextually Persuadable: Inference Dynamics in Large Language Models under Contextual Influence
Comments
Please log in to post a comment.