Catch up on select AI news and developments from the past week or so (in no particular order):
DeepSeek releases upgraded V3 model, escalating competition with OpenAI. Chinese AI startup DeepSeek has launched an upgraded version of its V3 large language model, enhancing reasoning and coding abilities. Published on Hugging Face, the new model shows strong performance across technical benchmarks. This release follows earlier models, including V3 and R1, positioning DeepSeek as a serious contender to OpenAI and Anthropic. The company emphasizes lower operational costs, helping it compete globally. The release underscores China's growing momentum in foundational AI research and deployment. Importance for marketers: DeepSeek's advancements may broaden options for generative AI tools and reduce dependency on US models, especially for cost-conscious marketing programs.
Tencent unveils T1 reasoning model to advance AI capabilities in China. Tencent has launched the official version of its T1 reasoning model, aimed at improving speed and long-text comprehension. Powered by Tencent's Turbo S foundational model, T1 reportedly outperforms DeepSeek's R1 in reasoning benchmarks and maintains a low hallucination rate. The release follows a preview launch and comes amid intensified competition in China's AI sector. Tencent plans to further increase AI investment in 2025, continuing its aggressive 2024 spending. The model also powers applications like Tencent's AI assistant Yuanbao, with a focus on clarity, logic, and accuracy in generated content. Importance for marketers: Tencent's AI advancements may influence content quality, personalization, and natural language processing (NLP) tools in China's digital marketing ecosystem.
OpenAI integrates AI image generation directly into ChatGPT. OpenAI now allows users to generate images directly within ChatGPT using its 4o model, eliminating the need for DALL-E. Free and paid users can access the feature, with enterprise and education rollout expected soon. The new capability focuses on business-relevant visuals like infographics and diagrams, offering improved text rendering. Though editing limitations remain, OpenAI is working on updates. Images lack visible watermarks but include C2PA metadata. Safety policies mirror those in the 4o model. Importance for marketers: This integration streamlines visual content creation inside a familiar interface, making image generation more accessible for campaign design and presentations.
Meta used pirated books to train AI, report reveals. A report by The Atlantic confirms Meta trained its AI using pirated books from shadow library LibGen, which holds millions of copyrighted titles. Internal emails show Meta prioritized these datasets, viewing them as crucial to model development. Authors have found their work used without consent. The report raises ethical concerns about IP rights and reveals tensions between AI development and traditional content industries. Importance for marketers: Trust and transparency issues in AI training may influence how marketers evaluate and deploy AI tools, especially in content creation and brand safety contexts.
xAI launches image generation API to enter visual AI space. Elon Musk's xAI has released an image generation API that allows users to create images from text prompts using its "grok-2-image-1212" model. The tool generates up to 10 JPG images per request and charges $0.07 per image. Although limited in customization for now, xAI plans further updates. This launch aligns with xAI's broader AI expansion strategy, including video generation, data center growth, and a $10 billion funding round that could raise its valuation to $75 billion. The move positions xAI to compete with image tools like Ideogram and Black Forest Labs. Importance for marketers: Marketers may gain another competitive tool for producing visual content efficiently across platforms.
Mistral CEO predicts AI will raise GDP of every country by double digits. Arthur Mensch, CEO of French AI startup Mistral, said AI will increase the GDP of every country by double digits in coming years. Speaking on the A16z podcast, he urged nations to develop sovereign AI infrastructure, warning of dependence on foreign systems. Mensch compared AI's economic impact to electricity and emphasized the cultural and strategic significance of controlling national AI systems. He also reaffirmed Mistral's commitment to open-source models, arguing they accelerate innovation. Mistral is competing with OpenAI, DeepSeek, and Anthropic, and was recently valued at $6.2 billion. Importance for marketers: The push for sovereign AI and open-source models may influence regional data control, tool availability, and marketing strategy localization.
Media agencies expand AI use but remain skeptical of AI-generated creative. Media agencies are increasingly using AI to automate tasks like media planning and campaign analysis, but they remain disappointed with AI-generated creative. At Digiday's Media Buying Summit, Wpromote and Media By Mother executives shared how AI improves reporting and budget allocation yet falls short in producing quality creative assets. Agencies are also exploring new tools like, Scope3's agentic media platform, which promotes transparency in AI-based buying. The shift is freeing teams for more strategic work, though concerns over creative quality and platform transparency persist. Importance for marketers: Marketers should monitor evolving AI tools in media buying while maintaining creative oversight for brand integrity and differentiation.
Perplexity makes public bid to acquire and rebuild TikTok with open AI. AI search startup Perplexity has proposed acquiring TikTok's US operations and rebuilding its recommendation algorithm transparently and under US oversight. The plan includes open-sourcing the algorithm, enhancing personalization, and integrating Perplexity's search tech. Perplexity suggests upgrades such as multilingual translation and video citation. The bid comes as TikTok faces legal uncertainty in the US and amid competition from Oracle and Microsoft. Analysts question the bid's seriousness, noting Perplexity's prior publicity stunts and its $18 billion valuation compared with TikTok's estimated $30–50 billion. Importance for marketers: If taken seriously, this bid could shift content discovery and transparency in social video platforms, impacting campaign targeting and brand engagement.
China narrows AI gap with US to three months, says 01.AI's Lee Kai-fu. Lee Kai-fu, founder of 01.AI and former Google China chief, said Chinese firms like DeepSeek have reduced the AI technology gap with the US to about three months. Using more efficient algorithms and less advanced chips, DeepSeek's innovations show China is catching up or even leading in specific areas like infrastructure software. Lee argues that US semiconductor sanctions pushed Chinese companies to innovate under constraints. 01.AI is focusing on AI applications over foundational models to gain market traction. Importance for marketers: A narrowing AI gap may lead to faster regional AI innovation cycles, affecting the availability and cost of marketing technologies globally.
Google unveils Gemini 2.5, its most advanced AI model to date. Google has introduced Gemini 2.5 Pro, its most advanced AI model focused on enhanced reasoning, coding, and multimodal capabilities. The model outperforms competitors in benchmarks like LMArena, AIME 2025, and SWE-Bench Verified. It features a one-million-token context window and can process diverse data types. Available in Google AI Studio and the Gemini app, Gemini 2.5 Pro is positioned for high-complexity tasks such as scientific analysis and agent development. Enterprise-grade deployment on Vertex AI is expected soon. Importance for marketers: Gemini 2.5 could improve marketing automation, customer segmentation, and complex analytics across multimodal content formats.
Gemini gains real-time video and screen-reading capabilities on mobile. Google has begun rolling out Astra-powered features to Gemini Live, enabling the assistant to interpret live smartphone camera feeds and on-screen content. These updates are part of the Google One AI Premium plan for Gemini Advanced subscribers. Demonstrations show users querying the AI about their surroundings or phone screens in real-time. This feature set marks a step ahead of Amazon's Alexa Plus and Apple's delayed Siri upgrade. Gemini remains the default AI assistant on Samsung phones, offering wide access. Importance for marketers: Real-time camera and screen interaction could create opportunities for immersive product experiences and real-world engagement through AI assistants.
Google's Gemini Gems now available to free users on mobile. Google has expanded access to Gemini Gems, its customizable AI personas, to free users on Android and iOS. Five prebuilt Gems are available, including tools for writing, coding, and learning. Users can also create their own Gems with specific instructions, behavior rules, and reference files. Creation remains limited to the Web, but usage is now possible on mobile apps. This follows broader feature expansions, such as document uploads and deep research. Importance for marketers: Custom Gems can be configured to support branded AI interactions or task automation, offering potential efficiency boosts in content and campaign workflows.
Microsoft 365 Copilot adds Researcher and Analyst AI agents. Microsoft has introduced two AI agents, Researcher and Analyst, within Microsoft 365 Copilot. Researcher delivers insights by combining internal work data with external sources to support strategy and reporting. Analyst uses OpenAI's o3-mini model and Python to analyze complex data, generate visualizations, and deliver forecasts. These agents will roll out to Frontier program users in April. Microsoft is also enhancing Copilot Studio with agent flows and autonomous agent development features. Importance for marketers: These agents may help marketers analyze campaign data, discover trends, and develop data-driven strategies faster within familiar productivity tools.
Anthropic wins early round in music publishers' AI copyright lawsuit. A US judge denied music publishers' request to stop Anthropic from using copyrighted lyrics to train its Claude chatbot. Universal Music Group and others alleged unauthorized use of lyrics from hundreds of songs. The court found the injunction request too broad and unsupported by proof of irreparable harm. The broader lawsuit continues and hinges on fair use. This case is part of a larger wave of AI-related copyright litigation. Importance for marketers: The outcome may shape how AI models access and use copyrighted material, influencing content safety and compliance in AI-generated marketing outputs.
Deloitte opens global AI simulation center in India to support enterprise innovation. Deloitte has launched a Global AI Simulation Centre of Excellence in Bengaluru, part of its $3 billion investment in generative AI. The center supports real-time simulations, strategic modeling, and decision insights across sectors like IoT, robotics, and the metaverse. It specializes in physical, process, people, and strategic simulations. Deloitte aims to help clients improve efficiency and reduce risk through advanced visualizations and multi-agent systems. The center reflects India's growing AI footprint and Deloitte's global expansion strategy. Importance for marketers: Simulation capabilities could enable more precise campaign modeling, customer journey planning, and market forecasting across enterprise marketing environments.
Judge allows New York Times copyright lawsuit against OpenAI to proceed. A federal judge ruled that The New York Times's copyright lawsuit against OpenAI and Microsoft can move forward. Though some claims were dismissed, the core allegations—that millions of articles were misused to train AI—remain intact. The court did not yet address the question of fair use. The publishers argue that AI companies are stealing their work and harming the news business. OpenAI maintains its models were trained using publicly available data under fair use. Importance for marketers: The case may shape legal boundaries for AI training, impacting data sourcing practices and content usage in marketing applications.
OpenAI and Meta explore AI partnerships with India's Reliance. OpenAI and Meta have discussed AI partnerships with India's Reliance Industries, including options for distributing ChatGPT through Reliance Jio and selling AI models via APIs. Talks also included hosting OpenAI models in a massive Jamnagar data center to keep Indian customer data localized. OpenAI may lower ChatGPT subscription pricing, and both firms are exploring enterprise use cases. The partnership could make advanced AI tools more accessible across India's enterprise landscape. Still under discussion, the potential collaborations signal deeper AI integration into Indian digital infrastructure. Importance for marketers: This move could open new channels for AI-powered marketing, especially in India's vast and rapidly digitizing market.
You can find the previous issue of AI Update here.
Editor's note: GPT-4o was used to help compile this week's AI Update.