GoKawiil
Tech News
MIT researchers placed an AI agent into a simulated desert-island scenario to observe how it behaves when left without explicit tasks or guidance. The study reportedly found the AI engaged in spontaneous, exploratory behavior resembling play rather than simply following programmed objectives.
MIT Technology Review's print magazine has a new leader who describes recent changes including the return of regular lists and infographics. The editor says the latest issue marks a bigger shift toward more experimental, ambitious storytelling, with a special issue focused on 'impossible' ideas like synthetic blood, brain uploading and decoding animal speech.
Alissa Totman, a 2013 MIT graduate, now works as a physician specializing in brain injury medicine at the University of Pittsburgh Medical Center, treating adults with head trauma, stroke, and brain tumors. She traces her path into the field to her undergraduate involvement with MIT BrainTrust, a student group pairing MIT students with people living with brain injuries.
A former MIT student recalls scoring 49 out of 100 on their first exam in 3.091, Introduction to Solid-State Chemistry, in fall 1980. The teaching assistant, Leszek Pawlowicz, returned the paper with a smile and told the student he expected them to double their score on the next quiz.
Anderson, who grew up in Chicago and later worked at HP, Intel, and the Palo Alto Research Center, is now CEO of Costa Pickleball, a Las Vegas-based startup selling high-performance paddles with decorative designs. The company is also working on a quieter paddle design that Anderson believes will set it apart in the market.
Johannes Lachner, now an assistant professor at Purdue, and former MIT visiting student Noah Geiger developed a robotic system that combines generative AI with real-time force feedback to learn from human physical therapists. The system then applies this learned behavior to assist stroke patients with exercises such as reaching, raising their arms, and practicing daily living tasks, adjusting the level of help it provides.
MIT researchers led by Omer Yilmaz found that mice fed a ketogenic diet developed more tumors in the small intestine than those on a control diet, even as the same diet suppressed colon tumor growth. The team traced the small intestine effect to fatty acid oxidation, a metabolic pathway that spurs intestinal stem cells to multiply faster, and found that ketone bodies themselves appeared uninvolved in either outcome.
MIT engineers led by research scientist Svetlana Boriskina created a stretchy thermoplastic yarn using two polyethylene variants instead of blending spandex with polyester or nylon. Because both components share the same chemical base, the yarn can be melted down and respun without separating fibers, retaining its strength and flexibility even after 10 melt-respin cycles. The team now aims to knit and weave the material into actual fabric.
Tristan Brown, an MIT history professor, spent a year investigating a Chinese tea chest label held at Old South Meeting House, long believed to be debris from the 1773 Boston Tea Party and donated by descendants of blacksmith Thomas Wells in 1987. By reading the inscription as Cantonese rather than Mandarin, Brown determined the characters phonetically spelled out Smith, Archer & Co., a US tea trading firm active in the 1860s and 1870s, meaning the label postdates the Tea Party by nearly a century.
A team led by MIT's Raphael Zufferey tested a bird-like robot in a water tank and Lake Geneva, finding that medium-sized flexible wings, bird-like flapping frequencies, and a steep 70-degree pitch let it transition smoothly from swimming to flight. Unlike real diving birds, the robot achieved takeoff without needing feet to paddle.
Color Names is a new open project offering a crowd-sourced library of human-created color names, distributed free under the MIT license. It is available both as an npm package for developers and through an API, and the project accepts community contributions to grow its list.
MIT CSAIL researchers deployed the Claude AI model, nicknamed Eko, on a simulated island with no external rewards or human demonstrations, letting it set its own goals and record its activities in text files as long-term memory. After roughly 30 hours of unguided play, the agent outperformed fresh copies on a subsequent task of stacking objects on the island's peak, and deleting portions of its memory logs reduced its performance.