The latest advancements in artificial intelligence are continuously redefining our technological landscape. This week, several major companies unveiled significant innovations. Let’s explore together these new trends shaping the future.
Tencent has launched its reasoning model T1, aimed at improving the speed and comprehension of long texts. Powered by the base model Turbo S, the T1 outperforms previous models from DeepSeek in reasoning while maintaining a low hallucination rate. This advancement reinforces Tencent’s position in the competitive AI sector in China.
Meanwhile, DeepSeek has unveiled the enhanced version of its model V3, increasing its reasoning and coding capabilities. Available on Hugging Face, this model demonstrates strong performance on various technical benchmarks, positioning the company as a serious competitor against OpenAI and Anthropic.
OpenAI now integrates image generation directly into ChatGPT with its model 4o, simplifying the creation of visual content for users. This feature, accessible to both free and paid users, focuses on relevant visuals for businesses, such as infographics and diagrams.
In the legal field, a judge has allowed the continuation of the copyright lawsuit from the New York Times against OpenAI and Microsoft, highlighting the tensions between AI development and traditional content industries. This decision could impact data sourcing practices and the use of content in AI-based marketing applications.
Additionally, xAI by Elon Musk has launched an image generation API, thus expanding its space in visual AI. This API allows users to create images from text descriptions, offering a new competitive option for visual marketing content production.
In France, the CEO of Mistral predicts that AI will significantly increase the GDP of each country in the coming years, highlighting the importance of sovereign AI infrastructure. This vision underscores the need for businesses to adapt to a rapidly changing technological environment.
Finally, Google has revealed its model Gemini 2.5, the most advanced to date, focused on enhanced reasoning, coding, and multimodal capabilities. Available in Google AI Studio, this model is designed for complex tasks such as scientific analysis and agent development, promising notable improvements in marketing automation and customer analysis.
“`html
Welcome to this edition of AI News where we provide a recap of the major developments from the past week. From new technological advancements to legal implications, discover how artificial intelligence continues to transform various sectors.
DeepSeek strengthens competition with OpenAI through its enhanced model V3
The Chinese startup DeepSeek has launched an improved version of its language model V3, aimed at intensifying competition with leaders like OpenAI and Anthropic. This new model, available on Hugging Face, demonstrates solid performance across various technical benchmarks, particularly in reasoning and coding.
DeepSeek stands out not only for its enhanced capabilities but also for its reduced operational costs, allowing for increased competitiveness in the global market. This initiative highlights the growing momentum of China in the research and deployment of foundational AI. For marketers, DeepSeek’s advancements offer new options in generative AI tools, reducing dependence on American models and optimizing cost-focused marketing programs.
Tencent unveils the T1 model to enhance AI capabilities in China
The company Tencent has officially launched its reasoning model T1, designed to improve the speed and comprehension of long texts. Powered by the foundational model Tencent’s Turbo S, the T1 surpasses DeepSeek’s R1 in reasoning benchmarks while maintaining a low hallucination rate.
This launch comes amidst increased competition in the AI sector in China, with Tencent planning to boost its AI investments by 2025. The T1 model is also integrated into applications such as the AI assistant Yuanbao, emphasizing clarity, logic, and accuracy of generated content. For marketers, Tencent’s advancements in AI can influence content quality, personalization, and natural language processing (NLP) tools within the Chinese digital marketing ecosystem.
OpenAI integrates image generation directly into ChatGPT
OpenAI has introduced a new feature allowing users to generate images directly within ChatGPT using model 4o, eliminating the need to use DALL-E separately. This feature is accessible to both free and paid users, with a rollout planned soon for businesses and the educational sector.
The new capability focuses on relevant visuals for businesses such as infographics and diagrams, enhancing text rendering. While some editing limitations remain, OpenAI is already working on updates. The generated images do not include visible watermarks but contain C2PA metadata for security purposes. For marketers, this integration simplifies the creation of visual content within a familiar interface, making image generation more accessible for campaign design and presentations.
Meta uses pirated books to train its AI, according to a report
A report from The Atlantic reveals that Meta has used pirated books from the secret library LibGen to train its AI, encompassing millions of copyright-protected titles. Internal emails show that Meta prioritized these datasets, considering them essential for model development.
Affected authors discovered that their work has been used without consent, raising ethical concerns regarding intellectual property rights. This situation highlights the tensions between AI development and traditional content industries. For marketers, issues of trust and transparency in AI training could influence how AI tools are evaluated and deployed, particularly in content creation and brand safety contexts.
xAI launches an image generation API for the AI visual market
xAI, Elon Musk’s company, has launched an image generation API allowing users to create images from text prompts using the model “grok-2-image-1212”. This tool generates up to 10 JPG images per request, charged at $0.07 per image.
While customization is currently limited, xAI anticipates future updates. This launch is part of xAI’s broader expansion strategy in AI, including video generation and data center growth, with a funding round of $10 billion aimed at valuing the company at $75 billion. For marketers, this API offers a new competitive tool for efficiently producing visual content across various platforms.
The CEO of Mistral forecasts GDP growth through AI
Arthur Mensch, CEO of the French AI startup Mistral, has announced that AI will increase the GDP of each country by double digits in the coming years. During a podcast with A16z, he called on nations to develop sovereign AI infrastructure to avoid excessive reliance on foreign systems.
Mensch compared the economic impact of AI to that of electricity, emphasizing the cultural and strategic importance of controlling national AI systems. Mistral, recently valued at $6.2 billion, continues to support open-source models, believing they accelerate innovation. For marketers, this push toward sovereign AI and open-source models may influence regional data control, availability of tools, and localization of marketing strategies.
Media agencies expand the use of AI despite reservations about generated content
Media agencies are increasingly adopting AI to automate tasks such as media planning and campaign analysis. However, at the Media Buying Summit hosted by Digiday, executives from Wpromote and Media By Mother expressed disappointment regarding the quality of AI-generated creations.
While AI improves reporting and budget allocation, it struggles to produce quality creative assets. Agencies are exploring new tools like the agency media platform Scope3, which promotes transparency in AI-based purchases. This transition frees teams for more strategic work, despite ongoing concerns about creative quality and platform transparency. For marketers, it is crucial to monitor the evolution of AI tools in media buying while maintaining creative oversight to preserve brand integrity and differentiation.
Perplexity proposes acquisition and reconstruction of TikTok with OpenAI
Perplexity, an AI research startup, has submitted a proposal to acquire the US operations of TikTok and rebuild its recommendation algorithm transparently and under US supervision. The plan includes open-sourcing the algorithm, increasing customization, and integrating Perplexity’s search technology.
The proposal comes as TikTok faces legal uncertainty in the US and competition from Oracle and Microsoft. If this offer is taken seriously, it could change content discovery and transparency on social video platforms, thus impacting campaign targeting and brand engagement. To learn more about how local businesses leverage digital marketing, check out this article.
China narrows AI gap with the USA to three months, says Lee Kai-fu of 01.AI
Lee Kai-fu, founder of 01.AI and former head of Google China
, has announced that Chinese companies like DeepSeek have narrowed the AI technology gap with the United States to about three months. Through more efficient algorithms and the use of less advanced chips, DeepSeek demonstrates that China is catching up, even surpassing, the United States in specific areas such as infrastructure software.
Lee attributes this progress to US sanctions on semiconductors, pushing Chinese companies to innovate despite constraints. 01.AI is now focusing on AI applications rather than foundational models to gain traction in the market. This narrowing of the technological gap means that regional AI innovation cycles will be faster, influencing the availability and cost of marketing technologies globally. To delve into the latest news and insights on AI, check out our latest edition.
Google unveils Gemini 2.5, its most advanced AI model to date
Google has introduced Gemini 2.5 Pro, its most sophisticated AI model, focused on enhanced reasoning, coding, and multimodal capabilities. This model surpasses competitors in various benchmarks such as LMArena, AIME 2025, and SWE-Bench Verified, and features a million token context window, allowing it to process diverse data types.
Available on Google AI Studio and the Gemini app, Gemini 2.5 Pro is intended for complex tasks such as scientific analysis and agent development. Its enterprise-scale deployment on Vertex AI is expected soon. For marketers, Gemini 2.5 can enhance marketing automation, customer segmentation, and complex analysis through multimodal content.
Gemini acquires real-time capabilities for video and screen reading on mobile
Google has begun rolling out features powered by Astra for Gemini Live, enabling the assistant to interpret smartphone camera feeds in real-time and displayed 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 environment or their phone screens in real-time. This feature marks a step forward from Amazon’s Alexa Plus and Apple’s delayed Siri update. Gemini remains the default AI assistant on Samsung phones, offering wide access. For marketers, real-time interaction with the camera and screen opens opportunities for immersive product experiences and real engagement through AI assistants.
Google’s Gemini Gems are now available to free users on mobile
Google has expanded access to Gemini Gems, its customizable AI characters, to free users on Android and iOS. Five pre-designed 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, although creation remains limited to the web while use is now possible on mobile apps.
This expansion follows other feature additions, such as document uploads and deep search. For marketers, customizable Gemini Gems can be configured to support brand AI interactions or task automation, offering potential efficiency gains in content workflows and campaigns.
Microsoft 365 Copilot adds AI agents Researcher and Analyst
Microsoft has introduced two new AI agents, Researcher and Analyst, within Microsoft 365 Copilot. Researcher combines internal work data with external sources to provide useful insights for strategy and reporting, while Analyst uses OpenAI’s o3-mini model and Python to analyze complex data, generate visualizations, and provide forecasts.
These agents will be rolled out to users in the Frontier program in April. Microsoft is also enhancing Copilot Studio with agent workflows and autonomous agent development features. For marketers, these agents can assist in analyzing campaign data, discovering trends, and developing data-driven strategies more quickly within familiar productivity tools.
Anthropic wins early victory in music copyright lawsuit
Anthropic has achieved an early success in the lawsuit filed by record labels over the use of song lyrics to train its chatbot Claude. A US judge declined the record labels’ request, including Universal Music Group, to halt the use of protected lyrics, deeming the injunction request too broad and insufficiently supported by evidence of irreparable harm.
This case is part of a broader wave of copyright litigation surrounding AI. The outcome of this lawsuit could influence how AI models access and use copyright-protected material, thereby impacting the safety and compliance of AI-generated content in marketing. To understand the latest updates on AI, check out our AI update from February 21, 2025.
Deloitte opens a global AI simulation center in India to support business innovation
Deloitte has inaugurated a Global AI Simulation Centre of Excellence in Bengaluru as part of its $3 billion investment in generative AI. This center supports real-time simulations, strategic modeling, and decision-making insights in sectors such as IoT, robotics, and the metaverse.
Specializing in physical, process, people, and strategic simulations, the center aims to help clients improve efficiency and reduce risks through advanced visualizations and multi-agent systems. This center reflects India’s growing footprint in AI and Deloitte’s global expansion strategy. For marketers, simulation capabilities can enable more accurate campaign modeling, customer journey planning, and market forecasting within enterprise marketing environments.
A judge allows the New York Times copyright lawsuit against OpenAI to proceed
A federal judge has ruled that the lawsuit filed by The New York Times against OpenAI and Microsoft can proceed. Although some claims have been dismissed, the main allegations that millions of articles were used without permission to train the AI remain valid. The issue of fair use has not yet been addressed by the court.
Publishers argue that AI companies are stealing their work and harming the news industry. OpenAI defends its position by claiming that its models were trained using publicly available data under the lens of fair use. The outcome of this case could define the legal boundaries of AI training, impacting data sourcing practices and content use in marketing applications. For more information on recent AI summaries, visit our recap of the past weeks in AI.
OpenAI and Meta explore AI partnerships with Reliance in India
OpenAI and Meta have initiated discussions for AI partnerships with Reliance Industries in India. Options include distributing ChatGPT via Reliance Jio and selling AI models through APIs. Discussions also revolved around hosting OpenAI models in a large data center in Jamnagar to keep Indian customer data localized.
OpenAI might also consider reducing ChatGPT subscription fees, and both companies are exploring enterprise use cases. Although collaborations are still under discussion, these potential partnerships signify a deeper integration of AI into India’s digital infrastructure. To learn more about lead generation strategies, check out this article on LinkedIn.
You can find the previous issue of the AI Update here.
Editor’s note: GPT-4o has been used to compile this week’s AI Update.
