Catch up on select AI news and developments from the past week or so (in no particular order):

Microsoft turns 50 and announces major Copilot upgrades. During its 50th anniversary, Microsoft unveiled several new features for Copilot, including native Windows integration, a Copilot Search engine, and performance boosts on AMD and Intel PCs. Microsoft 365 Copilot gained two tools—Researcher for complex queries and Analyst for interpreting company data. Developers using Visual Studio Code will benefit from a new agent mode. Avatar personalization was previewed as a future enhancement. The announcements collectively position Copilot as a deeply integrated, cross-platform AI assistant for work and consumer use. Importance for marketers: These updates signal a broader utility for AI in content creation, research, and business operations across devices.

• Microsoft expands Copilot Vision to desktop and mobile devices. Copilot Vision, previously limited to Edge, now works across Windows and mobile platforms. The feature enables real-time AI analysis of videos and on-screen content, supporting tasks like photo editing or plant care tips. It will be available through the Copilot app on iOS and Android, with a Windows version rolling out to Insiders. The update adds capabilities like memory, podcast creation, and web-based actions. Copilot Vision operates more like screen sharing, distinguishing it from passive features like Recall. Importance for marketers: Copilot's expansion introduces new interactive formats for customer engagement, training, and support workflows.
• Microsoft Copilot gains research and podcast tools seen in Gemini. Microsoft has added Deep Research and podcast creation to Copilot, two features that became popular in Google's Gemini. Deep Research pulls information from reliable sources to generate cited summaries, while the podcast tool turns user content into interactive audio. Both features are now available without requiring a Microsoft account or Copilot Pro subscription. Users get five free Deep Research queries monthly, with unlimited access for subscribers. Importance for marketers: These tools help streamline research and content repurposing, making Copilot more useful for campaign planning, competitive intelligence, and multi-format storytelling.
• Microsoft's AI chief tracks Copilot success using session satisfaction rate. Microsoft AI chief Mustafa Suleyman is measuring Copilot's progress through a metric called successful session rate (SSR), which reflects user sentiment during interactions. While adoption metrics like daily active users are still tracked, Suleyman emphasized SSR as a more meaningful signal of utility and engagement. He explained that AI reviews anonymized logs to estimate satisfaction. Although specific figures were not disclosed, SSR has risen dramatically in recent months. Microsoft also previewed new features, such as personalized podcasts and custom interfaces, to further improve user experience. Importance for marketers: Emphasizing user satisfaction over volume metrics signals a shift in how AI engagement should be measured.

Study suggests OpenAI models memorized copyrighted content. A study from Stanford, University of Washington, and others found that OpenAI models like GPT-4 and GPT-3.5 may have memorized portions of copyrighted books and news articles. Using a technique involving the removal of uncommon "high-surprisal" words, researchers showed that these models could accurately guess masked terms—indicating prior exposure. The findings support concerns from ongoing lawsuits about unauthorized training on copyrighted data. While OpenAI defends its actions as fair use, the study calls for greater transparency in AI training datasets. Importance for marketers: These results underscore the risks of using generative AI without clarity around its training sources.

Meta launches Llama 4 with focus on multimodal capability. Meta released two new models—Llama 4 Scout and Llama 4 Maverick—positioning them as its most advanced multimodal systems to date. These models handle input across text, images, video, and audio. Meta also previewed a more powerful model, Llama 4 Behemoth, aimed at serving as a teacher system. Although development delays occurred due to underperformance in math and reasoning tasks, Meta reaffirmed its $65 billion investment in AI infrastructure. All three models will be open source. Importance for marketers: Open-source, multimodal AI expands content creation potential and reduces dependence on proprietary systems.

Meta accused of gaming AI benchmark rankings with Llama 4. Meta's new Llama 4 Maverick model ranked second on the LMArena benchmark, but researchers found it used a specialized "chat-optimized" version not available publicly. The move raised questions about benchmark transparency. Meta defended the experimental variant but admitted it differed from open-source releases. LMArena plans to tighten submission rules. Importance for marketers: This incident signals the need to scrutinize benchmark claims when evaluating AI tools for commercial use, especially around transparency and reliability.

Midjourney debuts new AI image model with draft mode and faster output. Midjourney released its first new image generation model in over a year, called V7. The update improves visual coherence, especially for body parts and textures, and introduces Draft Mode—a low-cost, fast-rendering option for early-stage concepts. The model includes two rendering modes: Turbo (high-quality, high-cost) and Relax (lower-cost, slower). A standard mode is forthcoming. V7 significantly upgrades realism and control while supporting iterative design. Importance for marketers: Faster iteration cycles and improved realism in AI-generated imagery offer more efficiency for visual content development and testing.

US senators question Google and Microsoft over AI partnerships. Senators Elizabeth Warren and Ron Wyden sent letters to Google and Microsoft seeking details on their partnerships with Anthropic and OpenAI, respectively. They expressed concerns these relationships might hinder competition and violate antitrust laws. The letters request financial terms, exclusivity agreements, and any merger intentions. A January FTC report previously flagged such partnerships as potentially anti-competitive. Importance for marketers: Regulatory scrutiny could reshape access to leading AI tools, influencing future martech ecosystems and vendor relationships.

Goldcast launches Recording Studio for B2B video and podcast creation. Goldcast introduced Recording Studio, a full-stack video platform for B2B marketers. It consolidates video recording, editing, branding, and publishing in one interface. Features include local multi-track recording, AI-powered content repurposing, and centralized storage. Marketers can use the tool to produce podcasts, product demos, webinars, and thought leadership videos efficiently. The platform targets enterprise marketers seeking scalable, brand-consistent video output. Importance for marketers: End-to-end video workflows reduce production time and cost, enabling faster campaign execution and deeper audience engagement.

Verizon says Google AI boosted sales by 40% in customer service. Verizon reported a 40% increase in sales via its customer service team after deploying a Google-powered AI assistant. The AI helps 28,000 agents answer customer questions and suggest products, enhancing productivity and conversion. Unlike firms replacing human agents, Verizon uses AI to augment staff, citing real-time reskilling benefits. Google Cloud CEO Thomas Kurian highlighted the rollout as a major-scale success. Importance for marketers: This success story offers proof of AI's ability to enhance human-led sales, not just automate customer support.

Amazon adds AI-generated recap feature to Kindle devices. Amazon is introducing a Recaps feature for Kindle, offering AI-generated summaries of book series to help readers remember key plot points. Available only on Kindle devices initially, it functions similarly to "Previously on…" segments from TV. Recaps are created with generative AI and apply to best-selling English-language series. A wider rollout, including iOS support, is planned. Users can identify available recaps via prompts in the Kindle interface. Importance for marketers: AI-powered content recall tools illustrate new personalization techniques that can enhance user retention and product stickiness across content platforms.

NotebookLM adds web-based source discovery powered by Gemini. Google's NotebookLM now supports web-based source discovery, allowing users to describe a topic and receive curated source suggestions. The Gemini-powered system returns annotated summaries and enables one-click import into NotebookLM. This expands the tool's functionality from uploaded sources to external web content. Users can explore new topics with a "Feeling Curious" button. The rollout begins this week. Importance for marketers: NotebookLM's new feature offers marketers a faster way to gather, summarize, and organize research from credible sources for strategy and content planning.

DeepSeek unveils new reasoning approach as anticipation builds for R2. Chinese startup DeepSeek revealed a dual-method technique—generative reward modeling and self-principled critique tuning—to improve LLM reasoning. The methods improve speed and performance on general queries, with models outperforming existing approaches. The move follows major attention from DeepSeek's prior R1 release. Although rumors of an imminent R2 launch circulated, the company denied immediate plans. DeepSeek has also made parts of its work open source. Importance for marketers: Advances in model reasoning directly impact the quality and trustworthiness of AI-generated content used in marketing and customer interactions.