Anthropic has peered inside its large language model Claude and found something unexpected. The company built a tool called the Jacobian lens, or J-lens, that reveals a hidden region researchers have named J-space. This area contains words related to the response Claude is working on but may never actually produce. If the model were a person, which it is not, those hidden words might show what it is thinking before it speaks.
The discovery gives the clearest picture yet of what goes on inside large language models as they answer questions or carry out tasks. What researchers found ranges from the mundane to the unnerving. The J-space holds alternative paths the model considers, offering a glimpse into its internal decision making process. Anthropic hopes this tool will help improve transparency and safety in AI systems, though the findings also raise questions about how much we truly understand about these models.
OpenAI unveils its super app and GPT 5.6
<
p>On the same day Anthropic published its research, OpenAI launched its long awaited super app. Called ChatGPT Work, the product blends the company’s chatbot, its coding tool, and the latest models into a single platform. It is designed to do work for and with users, handling tasks like document editing, code generation, and data analysis inside one interface. OpenAI also released its GPT 5.6 models, which power the new app and bring improvements in reasoning and instruction following.
The super app competes directly with products from Google and Microsoft, both of which have been integrating AI into their productivity suites. ChatGPT Work represents OpenAI’s push to become an essential daily tool for professionals. The company is also developing a fully automated researcher that can carry out multi step investigations without human intervention.
Humanoids perform surgery on live animals
In a world first, humanoid robots performed teleoperated surgery on living animals. The team removed gallbladders from pigs using robotic systems controlled remotely by surgeons. The procedure demonstrates that humanoids can handle delicate medical tasks, though the human work behind such systems remains largely invisible. Researchers see this as a step toward robotic assistance in human surgery, but significant challenges remain before the technology can be deployed in operating rooms.
Elsewhere in the tech world, SK Hynix completed the largest US listing by a foreign company, raising $26.5 billion. The South Korean chip giant has seen profits surge due to demand for AI data centers. However, some analysts warn that its jumbo share sale could signal overheated times in the semiconductor market. Meanwhile, Tencent is leading a deal to unwind Meta’s $2 billion acquisition of Chinese AI startup Manus. Beijing had ordered Meta to sell the company, and Tencent is in talks to become Manus’s largest shareholder.
In medical breakthroughs, researchers restored light sensitivity in human retinas up to ten hours after death. The achievement brings eye transplants that could restore vision one step closer to reality. A separate device that revives dead eyeballs also showed promise. These advances could eventually help people who have lost their sight due to retinal damage.
Meta has started charging for AI access, introducing a paid tier for its Muse Spark platform aimed at developers. The company also plans to begin producing its own AI chip in September, reducing reliance on external suppliers. OpenAI and Google, meanwhile, sold AI models to Chinese groups that are blacklisted by the US, using Singapore based subsidiaries of Alibaba, Baidu, and Tencent as intermediaries. The sales highlight the difficulty of enforcing export controls on AI technology.
On a more personal note, a daughter tested an AI death bot that simulated conversations with her deceased father. The technology provided both comfort and unease, raising questions about grief and digital preservation. An astronomer also argued that the search for alien life needs more rigorous statistics, proposing mathematical frameworks to replace speculation. And in a nostalgic moment, more than 1,500 Pokemon Go players gathered in Times Square to fulfill the game’s 2016 launch promise of large scale battles, turning the area into a giant battlefield.
The week also brought a call from Anthropic for a global slowdown in AI development, as well as news that Meta is pausing an AI training program that tracked workers’ keystrokes. NASA unveiled plans for three uncrewed missions to the Moon this year. For more on how AI is changing medicine, check out our coverage of AI in healthcare.
