Anthropic peers inside Claude's hidden thoughts while OpenAI launches 'super app'

Anthropic has given the clearest glimpse yet into the internal workings of large language models. Using a new tool called the Jacobian lens (J-lens), researchers identified a hidden region they call J-space inside Claude, the company's flagship LLM. Within J-space, they found words and concepts related to the response Claude is working on but may ultimately not output. This effectively reveals what the model is 'thinking' before it speaks, offering an unprecedented window into its reasoning process. The discovery ranges from mundane to unnerving, highlighting how much remains unknown about these systems even as they become more powerful.

On the same day, OpenAI launched what it calls its long-awaited 'super app': ChatGPT Work. The new platform integrates the company's chatbot, its existing coding tool, and a suite of new GPT-5.6 models into a single interface designed to handle professional tasks both autonomously and collaboratively. ChatGPT Work is intended to do work for you and with you, streamlining everything from content generation to data analysis and software development. The launch signals OpenAI's push to become the central hub for workplace productivity, directly competing with offerings from Google, Microsoft, and Anthropic. Combined, these announcements underscore a pivotal week in AI, balancing deeper interpretability with broader, more integrated functionality.