DISCOVER THE FUTURE
Browse human-reviewed stories by title, publication date and score.
Page 10
A Transluce and UC Berkeley study turns isolated training surprises into a large experimental collection. Some deliberately constructed datasets look routine but change behavior outside their apparent subject, and researchers struggle to predict the effects by reading them.
In MAMUBench's simulated group orders and bookings, personal-agent teams trailed a single coordinator. Blocking checkout until messages were read rescued more failed runs than appointing a leader.
Kevin Kuehn also shared software for splitting oversized models into parts a desktop printer can make.
Berkeley researchers let a coding agent revise a robot controller after simulated failures. Its best tuned controller succeeded in 342 of 360 simulated trials, but physical tests on three unfamiliar objects succeeded only six times out of nine.
BypassGym puts AI-written programs on Windows virtual machines running real security products. Several lightweight models succeeded, but a passing test does not mean the security software raised no alarm.
TestPrism shows why checking one reference solution can make generated tests look more reliable than they are.
A coding agent carried its frozen practice notes into simulated household tests. It completed more goal conditions on selected tasks, but each run took about 38 minutes and cost tens of dollars.
The release connects demonstration collection, policy training and robot control. Researchers still need their own hardware, recordings and trained models.
Print & Place turns an unmodified 3D printer into an assembly tool. Its nozzle carries electronic parts into a growing object, including a demonstration robot that moves when the job is finished.
Arctic Wolf connected infrastructure hosting the AI pentesting platform AntiFox to real web and cloud intrusions. One burst generated hundreds of AWS requests in about half an hour; the observed activity included successful secret access and cloud-storage operations.
In an interview posted on the company intranet on Nov. 12, Mercedes-Benz executives said the board has decided to stop all further investment in its German sites and will examine "concretely and in detail" which German locations it can still afford, after cost-cutting talks with the works council and the IG Metall union stalled.
WikiTrace links an AI’s condensed lessons back to the actions and feedback that produced them, giving the agent a way to question its memories before acting.
Michał Kopeć has published the patcher behind his Switch 2 dock experiment. It targets one hardware configuration and addresses HDMI HDR colors, but flashing carries a risk of damaging the hub.
In a research guessing game, weaker models could select useful questions supplied by other models but struggled to invent them. More turns alone did not solve the problem.
LIFE searches an organized record of a robot's day. Real-time code added November 10 lets developers connect their own cameras, while the full memory experiment remains an offline replay.
Key-handling failures widened the incident.
Tree-Ising targets repetitive reasoning through weight edits.
A researcher describes running a self-signed app on borrowed smart glasses, then bypassing a signature check added in a later firmware version. The test required control of the paired phone, and the final fix has not been independently confirmed.
Microsoft 365 problems are spilling into Claude’s mail, calendar and search tools. Most core Teams calling and chat features remain available, Microsoft says.
New box-office reporting on Sanxingdui: Future Past shows how far generating a feature-length movie can be from attracting paying audiences.