Claude Has a Space for Its Thoughts. Whether Anyone Is Home Is Another Question.
Anthropic's interpretability team found a small workspace inside Claude, the J-space, that holds thoughts the model can report, steer and reason with before it speaks. It matches five signatures of conscious access in human brains, and nobody built it: it emerged during training. Anthropic claims access, not experience, and the distinction is where the debate now lives.
Read full story →Three States Are Quietly Writing America's AI Rulebook
Illinois just became the third state to pass a frontier AI safety law, adding a first-in-the-nation annual audit that even OpenAI and Anthropic endorsed. With California and New York, a bloc holding 40 percent of the US AI market is setting national rules while Congress stalls. The catch: courts and the Justice Department are already circling.
Read full story →A Humanoid Robot Just Removed a Gallbladder
UC San Diego researchers report the first surgery performed by a general-purpose humanoid, nicknamed Surgie, published in Nature. It is a real milestone with a real catch: a human surgeon steered every motion. The pitch is cheap, portable surgery for places that have none, from rural hospitals to battlefields.
Read full story →Physical AI's Real Bottleneck Isn't Chips. It's Data.
Twenty billion dollars has flowed into robots that act in the real world, and this week Citi's analysts kept naming the one constraint money can't fix: proprietary data from robots actually doing the work. The surest profits, for now, go to the picks-and-shovels sellers like lidar maker RoboSense, up 510 percent.
Read full story →