OpenAI has introduced a new AI agent called “deep research,” which is designed for comprehensive research tasks. Built on the recently launched o3 model, this agent can conduct complex internet searches and analyses in minutes instead of hours. In a benchmark test called “Humanity’s Last Exam,” deep research answered 26.6% of scientific questions correctly, outperforming previous models like GPT-4o and o3-mini-high.
Currently, deep research is available only to Pro users with paid accounts and is limited to text outputs, with plans to include images and graphics in the future. The research processing time ranges from 5 to 30 minutes. OpenAI emphasizes the need for human oversight (Human-in-the-Loop) due to potential inaccuracies, although these are reportedly less frequent than in other models. As of now, deep research is not available in the EU.
OpenAI has also announced the release of o3-mini, a new reasoning model available for free ChatGPT users and API access. This model excels in scientific, mathematical, and programming tasks. It offers three “Reasoning-Effort” options: low, medium, and high, allowing developers to balance speed and accuracy according to their needs. The highest level is recommended for programming and logical tasks. Paying ChatGPT users can choose between two options, such as o3-mini-high. The o3-mini model replaces the previous o1-mini for free users, while Plus and Team users see their daily message limit increased from 50 to 150. GitHub has integrated o3-mini into its Copilot and GitHub Models tools.
Meanwhile, the Chinese AI chatbot DeepSeek has shown significant user growth but still lags behind the market leader, ChatGPT. On January 24, the DeepSeek website saw 6.2 million daily visits, a substantial increase from 300,000 in December 2024. Its Android app had 1.64 million active users on the same day. However, ChatGPT had 117.5 million website visits on that day, 19 times more than DeepSeek. In mobile apps, ChatGPT had about 50 times more daily active users.
The Authors Guild, the largest US writers’ organization, has introduced a new certification for books that are verifiably human-authored. This initiative comes in response to the growing presence of AI-generated books in online stores. The “Human Authored” logo allows minimal AI use for grammar and spell-checking. Certified books will be listed in a public database. The organization stresses that this is not a rejection of new technologies but a move for transparency and recognition of the unique human aspects of storytelling. Currently, only Authors Guild members can apply for the seal, with plans to extend it to other authors.
Microsoft is expanding the AI capabilities of Windows 11 to more apps like Paint. Beta testers have access to a new preview version of Windows 11 that includes a Copilot button in Paint for additional AI functions. Currently, Paint users on Windows 11 only have access to the “Image Creator,” which generates image content based on AI prompts. The new Copilot button will offer more AI features, such as Cocreator, Image Creator, a generative eraser, and AI-assisted background removal. However, using these new AI features in Paint requires a subscription.
Google’s phone AI, Duplex, is back to automate appointment scheduling. This feature, known as “Ask for Me,” allows users to specify the service they need and for what purpose, with Google’s assistant handling the rest. A similar function was introduced nearly seven years ago with the phone AI “Duplex,” designed to make appointments on behalf of the user.
