AI News, February 20, 2025

OpenAI has reached 400 million weekly active users, marking significant growth since December 2024. OpenAI has announced that it now serves 400 million users every week, reflecting a substantial increase from the 300 million users it had in December 2024. This growth underscores OpenAI's expanding presence as a consumer company, particularly with its AI chatbot, ChatGPT. Additionally, OpenAI's enterprise business has seen notable growth, with 2 million paying users, doubling since September 2024. The company's developer traffic has also doubled in the past six months, highlighting its thriving business amidst new competition from China's DeepSeek.

Source: https://techcrunch.com/2025/02/20/openai-now-serves-400m-users-every-week/

OpenAI's user base has grown significantly, with 400 million weekly active users and doubling enterprise users. OpenAI's Chief Operating Officer, Brad Lightcap, has revealed that the company has reached a milestone of 400 million weekly active users. This growth is attributed to the increasing utility and familiarity of AI, particularly with ChatGPT. OpenAI's enterprise business has also seen substantial growth, with the number of paying users doubling to 2 million since September 2024. Despite new competition from DeepSeek, OpenAI remains focused on its strategic plans, viewing the emergence of competitors as a testament to AI's mainstream impact.

Source: https://www.pymnts.com/news/artificial-intelligence/2025/openai-chief-operating-officer-brad-lightcap-says-company-reaches-400-million-weekly-active-users/

OpenAI partners with Schibsted Media Group and introduces data residency in Europe. OpenAI has recently partnered with Schibsted Media Group, further expanding its reach in the media sector. Additionally, the company has introduced data residency in Europe, enhancing data management capabilities for European users. This move aligns with OpenAI's efforts to strengthen its global presence while ensuring compliance with regional data regulations. The partnership and data residency initiative highlight OpenAI's commitment to collaborative innovation and user data security.

Source: https://openai.com/newsroom/

Google's AI co-scientist aims to speed up scientific research by generating novel hypotheses and refining experiments. Google has introduced an AI co-scientist built on its Gemini 2.0 platform, designed to accelerate scientific research by generating novel hypotheses, drafting proposals, and refining experiments. This system is intended to collaborate with scientists, helping them uncover new knowledge and formulate research hypotheses. Early trials have shown mixed results, with the system outperforming state-of-the-art models in some tests but requiring further evaluation for broader applicability.

Source: https://www.computerworld.com/article/3829135/googles-new-ai-co-scientist-aims-to-speed-up-the-scientific-discovery-process.html

Google shared its latest AI advancements, including Gemini 2.0 Flash and AI tools for education and retail. Google highlighted its recent AI advancements in January 2025, focusing on integrating AI benefits into various fields. Key updates included the performance upgrade of Gemini 2.0 Flash, enhancing the Gemini app's capabilities for tasks like brainstorming and learning. Additionally, Google showcased AI tools for education and retail, such as the Automotive AI Agent for automakers like Mercedes-Benz. These developments aim to improve efficiency and user experiences across different sectors.

Source: https://blog.google/technology/ai/google-ai-updates-january-2025/

Google makes Gemini 2.0 models generally available, introducing new versions like Gemini 2.0 Flash-Lite. Google has announced that Gemini 2.0 models are now available to everyone, including updates like Gemini 2.0 Flash and the introduction of Gemini 2.0 Flash-Lite. These models offer improved performance and cost efficiency, with Gemini 2.0 Flash being a workhorse for high-volume tasks and Gemini 2.0 Flash-Lite providing better quality at the same speed and cost as its predecessors. Developers can access these models through the Gemini API in Google AI Studio and Vertex AI.

Source: https://blog.google/technology/google-deepmind/gemini-model-updates-february-2025/

Perplexity AI has launched its Deep Research tool, offering a competitive alternative to OpenAI and Google in AI-driven research. Perplexity AI has introduced its Deep Research tool, marking a significant step in AI-powered research. This tool synthesizes information from diverse sources to generate comprehensive reports quickly and accurately. It achieves a notable 26.6% accuracy on the Humanity's Last Exam benchmark, outperforming some competitors. Available under a freemium model, it offers free access with usage limits or unlimited use for $20/month, making it more accessible than premium offerings from Google and OpenAI.

Source: https://opentools.ai/news/perplexity-ai-unveils-deep-research-tool-a-new-era-in-ai-powered-insights

Perplexity Deep Research combines autonomous reasoning with rapid processing to deliver exhaustive reports on specialized topics. Perplexity Deep Research is a cutting-edge tool launched by Perplexity AI in February 2025. It employs a proprietary framework called test time compute (TTC) expansion to systematically explore complex topics. Unlike traditional search engines, it performs dozens of searches, evaluates hundreds of sources, and synthesizes findings through probabilistic reasoning models. This approach allows it to reconcile contradictory information and prioritize authoritative sources, making it faster than similar tools from Google and OpenAI.

Source: https://www.zdnet.com/article/what-is-perplexity-deep-research-and-how-do-you-use-it/

Forbes and Wired have accused Perplexity of plagiarism in its beta Perplexity Pages feature. Perplexity AI has faced accusations of plagiarism and unethical web scraping from news outlets like Forbes and Wired. The controversy centers around Perplexity's beta feature, Perplexity Pages, which allegedly reproduces content from original sources without proper citation. While Perplexity has committed to improving source citations, the issue highlights challenges in AI content generation, including the potential for hallucinated links and the nuances of fair use.

Source: https://techcrunch.com/2024/07/02/news-outlets-are-accusing-perplexity-of-plagiarism-and-unethical-web-scraping/

Anthropic is exploring the use of its AI model Claude to improve UK public services. Anthropic, an AI safety and research company, has signed a memorandum of understanding with the UK Secretary of State to explore integrating its AI model Claude into public services. This collaboration aims to enhance the efficiency and effectiveness of public services by leveraging Claude's capabilities. The move highlights Anthropic's commitment to applying AI in real-world applications, particularly in governance and service delivery.

Source: https://rivalsense.co/intel/anthropic-latest-news-updates-feb-16-2025-release/

Anthropic is set to release a hybrid AI model that combines deep reasoning with fast responses. Anthropic is gearing up to launch a new hybrid AI model that can switch between deep reasoning and fast responses. This model is expected to outperform OpenAI's o3-mini-high model on some programming tasks and will include a sliding scale for cost control. The hybrid approach aims to provide developers with flexibility in managing computational resources while leveraging advanced AI capabilities.

Source: https://techcrunch.com/2025/02/13/anthropics-next-major-ai-model-could-arrive-within-weeks/

Anthropic's Claude model can generate structured outputs like JSON, though it requires specific prompting techniques. Anthropic's AI model, Claude, has shown capabilities in generating structured outputs such as JSON, although it often requires additional techniques like 'tool calls' to ensure consistency. While Claude can interpret field descriptions effectively, it does not directly support Pydantic models like OpenAI's GPT-4o. This highlights the ongoing efforts to improve the structured output capabilities of commercial LLMs.

Source: https://towardsdatascience.com/enforcing-json-outputs-in-commercial-llms-3db590b9b3c8/

Microsoft has launched an AI Hub to promote apps that utilize neural processing units (NPUs) in Co-Pilot+ PCs. Microsoft has introduced an upgraded AI Hub to highlight applications that support neural processing units (NPUs) in Co-Pilot+ PCs. This move aims to encourage more developers to create NPU-centric apps, enhancing the utility of AI PCs. The NPU is designed to efficiently handle AI-intensive tasks like image recognition and content creation, offloading these from the CPU and GPU for faster system performance. Despite the initial lack of NPU-supported apps, Microsoft and partners like Dell are working to increase developer engagement and expand the app catalog.

Source: https://www.computerworld.com/article/3829129/microsofts-new-ai-hub-highlights-apps-that-support-npus.html

Microsoft highlighted AI innovations transforming industries in the UAE. Microsoft recently showcased groundbreaking AI applications in the UAE, emphasizing the country's role as a hub for digital innovation. The event highlighted Microsoft's AI solutions empowering businesses across sectors like education, finance, and energy. This collaboration with forward-thinking organizations aims to drive AI transformation and economic growth in the region. The event brought together key stakeholders from both public and private sectors to explore AI's transformative potential.

Source: https://news.microsoft.com/en-xm/2025/02/06/microsoft-showcases-remarkable-innovations-accelerating-ai-transformation-and-growth-across-the-uae/

Azure OpenAI Service allows setting response formats to return valid JSON objects. Microsoft's Azure OpenAI Service has been enhanced with the ability to set response formats to return valid JSON objects. This feature, known as JSON mode, requires specific instructions in the system message to ensure the model outputs JSON correctly. It is crucial for structured data exchange and can be particularly useful in applications requiring precise data formatting. This capability simplifies the integration of AI outputs into various software systems.

Source: https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/json-mode