News

Chinese AI startup DeepSeek has quietly launched an upgraded version of its R1 reasoning model, DeepSeek-R1-0528, on the open-source platform Hugging Face. Emerging around May 29, 2025, this release highlights the intensifying global competition in artificial intelligence, particularly in complex reasoning tasks where DeepSeek-R1-0528 demonstrates performance comparable to industry leaders OpenAI and Google.

The upgraded model reportedly achieves “significantly enhanced depth of reasoning and inference capabilities” through increased computational resources and algorithmic optimization during post-training. Benchmark results from platforms like Artificial Analysis indicate that DeepSeek-R1-0528 has reached parity with OpenAI’s o3 and Google’s Gemini 2.5 Pro models across key evaluations. In the AIME 2025 test, it scored 87.5%, close to OpenAI-o3’s 88.9% and surpassing Gemini 2.5 Pro’s 83.0%. On the LiveCodeBench coding benchmark, it matched Gemini 2.5 Pro at 77% and closely approached OpenAI o3’s 78%. Similarly, it recorded 85% on the MMLU-Pro reasoning and general knowledge benchmark, aligning with both competitors.

DeepSeek’s development strategy prioritizes architectural efficiency over raw computing power. A previous model, DeepSeek V3, achieved high performance using significantly fewer GPUs—2048 NVIDIA H800s—than expected for its capability level, a point underscored by former OpenAI researcher Andrej Karpathy. This emphasis on efficiency could drive both rapid advancements and cost-effectiveness.

The open-source availability of DeepSeek’s models has enabled widespread adoption across various sectors in China. Reports suggest their use in research for advanced warplanes, integration into BMW vehicles, legal document drafting by courtroom officials, and aiding agencies in locating missing persons. This swift transition from model deployment to practical applications underscores the growing impact of accessible, high-performance AI.

DeepSeek’s progress and the competitive R1 model highlight China’s continued AI advancements, despite U.S. export restrictions on advanced chips. The company’s approach suggests that breakthroughs in AI rely not only on extensive computational clusters but also on algorithmic innovation and efficient model design. For consumers and the broader tech industry, this signifies an increasingly diverse and competitive global AI landscape, potentially accelerating innovation and expanding choices in AI-driven tools and services.

Elon Musk’s AI venture, xAI, is taking a major step toward expanding the reach of its chatbot, Grok, by integrating it with Telegram, one of the world’s largest messaging platforms. The deal, reportedly valued at $300 million over one year, is structured to offer significant financial and strategic advantages to both parties while positioning Grok for broader consumer adoption.

Under the terms of the agreement, Telegram will receive a combination of cash and equity from xAI, along with 50% of revenue from Grok Premium and Premium+ subscriptions sold through the Telegram platform. Telegram founder Pavel Durov confirmed the partnership, emphasizing the mutual benefits and strategic alignment between the two companies.

For xAI, the deal opens access to Telegram’s user base of over one billion monthly active users—a significant leap from the estimated 500 million monthly users of X (formerly Twitter), where Grok is currently deployed. By embedding Grok into Telegram, xAI aims to place its AI assistant directly within users’ daily communication tools, a move that could drastically increase engagement and exposure.

Data Access and AI Training Potential

While the agreement signals a clear expansion in Grok’s distribution, it also raises questions about data usage and training rights. On platform X, Grok is trained in part using public user content—outside of jurisdictions like the EU. Whether a similar arrangement applies to Telegram remains unclear.

Telegram hosts a large number of public Channels, used for broadcasting messages to wide audiences. These Channels could, in theory, provide a valuable source of structured and domain-specific text data—particularly appealing at a time when high-quality training data for large language models is increasingly scarce and fragmented. If xAI gains access to such data, the deal could serve both distribution and model improvement goals simultaneously.

Distribution as the Next Frontier in AI

The Telegram integration reflects a broader trend in the AI space: distribution and embedded access are becoming as critical as model quality. While developing state-of-the-art AI systems remains essential, success increasingly depends on making those systems accessible where users already are. With its integration into a widely used messaging app, Grok follows a path similar to AI assistants from Meta, Google, and OpenAI, which are being embedded in chat apps, productivity tools, and hardware.

This also points to a future where messaging platforms could become key revenue channels for AI services. With Grok Premium subscriptions now accessible via Telegram, xAI taps into not only a massive user base but also a new monetization layer, shared directly with the platform hosting the interaction.

For consumers, the move means another AI assistant—this time with Elon Musk’s branding and capabilities—will become natively available within Telegram, offering potential benefits in productivity, conversation, and content generation. For the industry, it underscores the strategic importance of pairing AI technologies with high-traffic, high-frequency user platforms in order to win the race for user engagement.

Reed Hastings, co-founder and former CEO of Netflix, has joined the board of directors at Anthropic, an AI safety and research company known for its focus on building reliable and aligned AI systems. The appointment was made by Anthropic’s Long-Term Benefit Trust around May 28, 2025, and reflects the organization’s growing emphasis on ethical governance and societal impact at the highest level of oversight.

Hastings brings decades of leadership experience in building and scaling global technology platforms. In addition to founding Netflix, he has held board positions at leading technology firms including Facebook (Meta) and Microsoft. According to Buddy Shah, Chair of the Long-Term Benefit Trust, Hastings was selected for his “impressive leadership experience, deep philanthropic work, and commitment to addressing AI’s societal challenges,” making him “uniquely qualified to guide Anthropic at this critical juncture in AI development.”

The move is consistent with Anthropic’s mission to build interpretable and steerable AI systems that benefit humanity. Hastings has increasingly focused his attention on the intersection of technology and ethics. Notably, he recently donated $50 million to Bowdoin College to launch a research initiative on AI and Humanity, aimed at understanding how AI is transforming work, relationships, and education while developing responsible frameworks for its use—issues that closely mirror Anthropic’s research agenda.

In a public statement, Hastings expressed support for Anthropic’s dual commitment to optimism and caution in AI development:

“Anthropic is very optimistic about the AI benefits for humanity, but is also very aware of the economic, social, and safety challenges. I’m joining Anthropic’s board because I believe in their approach to AI development, and to help humanity progress.”

Daniela Amodei, President of Anthropic, welcomed Hastings by highlighting his broader awareness of the societal responsibilities borne by tech companies:

“His focus on the human impact of technology—whether at Netflix or through his global health and education initiatives—makes him an ideal addition to our board as we continue building AI that helps rather than harms.”

The appointment of such a high-profile figure suggests that Anthropic is committed not only to technical excellence but also to embedding ethical foresight into its governance structures. Hastings’ experience straddling disruptive innovation and philanthropic engagement adds a distinctive perspective to the company’s board at a time when public and regulatory scrutiny of AI development is intensifying.

His presence may also influence broader industry norms, particularly around the composition and function of AI governance boards. As companies like Anthropic, OpenAI, and Google DeepMind expand the reach of advanced AI, integrating leadership that understands both the technological potential and societal responsibility may become a defining factor in building public trust—and long-term viability.

Meta’s AI assistant has reached a major milestone, now serving over 1 billion monthly active users across the company’s extensive ecosystem of applications. The announcement came during Meta’s annual shareholders’ meeting, where CEO Mark Zuckerberg emphasized the company’s vision to position Meta AI as the leading personalized AI assistant.

Integrated into core platforms like Facebook, Instagram, WhatsApp, and Messenger, Meta AI is also available through Ray-Ban Meta smart glasses and Meta Quest headsets. This wide distribution is seen as a key driver of its rapid adoption. According to Zuckerberg, Meta’s strategic focus for 2025 is to “deepen the experience and make Meta AI the leading personal AI with an emphasis on personalization, voice conversations, and entertainment.”

Despite the impressive user base, it is acknowledged that some interactions with Meta AI may be unintentional—an effect of the assistant’s pervasive presence across Meta’s products. However, Chief Financial Officer Susan Li noted that WhatsApp is seeing the fastest growth in AI usage, with users actively engaging in chats, asking questions, and generating images. This points to more intentional and value-driven use cases emerging, particularly within messaging environments.

Looking ahead, Meta is actively exploring monetization strategies for its AI assistant. Potential avenues include paid recommendations, premium subscriptions, or enhanced access to computing resources. These monetization options were previously hinted at with the April launch of the standalone Meta AI app, which may serve as a testing ground for future offerings. “Having 1 billion monthly active users is no longer an aspirational target—it is now our current reality,” Zuckerberg stated, underlining the platform’s scale and influence.

The rapid ascent of Meta AI underscores the effectiveness of embedding emerging technologies into high-traffic platforms. Personalization, conversational capabilities, and entertainment use cases—particularly on apps like WhatsApp—are proving to be key engagement drivers. As Meta moves toward monetization, the platform is expected to introduce tiered AI experiences, aligning with broader industry trends seen at Google, OpenAI, and Apple.

With this shift, Meta is not only cementing its place in the AI assistant race but also accelerating competition across the tech landscape, where the future of personalized AI is quickly becoming a central battleground for user attention and revenue.

Anthropic is expanding the capabilities of its AI assistant, Claude, with two key updates announced on May 28, 2025. Web search is now available to all users free of charge, and a new voice interaction mode has entered beta testing on mobile devices.

Previously restricted to premium subscribers, Claude’s web search functionality now allows all users to access real-time information from the internet for queries, trend analysis, and research tasks. This broader availability is expected to increase adoption and provide valuable interaction data for refining Anthropic’s AI models.

At the same time, Anthropic is introducing a beta version of voice mode for its iOS and Android apps. This feature enables spoken conversations with Claude, supporting English and offering a tap-to-interact function for dynamic conversations. Users can seamlessly switch between voice and text input within a session.

Free users will have access to basic voice interactions, with a daily limit of 20–30 messages, while more advanced integrations—such as voice control for Google Docs, Gmail, Calendar, and web searches—will be reserved for paying subscribers. This tiered approach balances accessibility with premium feature incentives.

These enhancements follow the recent launch of the Claude 4 model family, including Opus 4 for complex tasks and Sonnet 4 for faster performance.

User Benefits

Free users now gain real-time web search access, significantly improving Claude’s utility. Voice mode introduces a more natural way to interact, making the AI more adaptable for various tasks and environments.