In June, OpenAI unveiled the largest language model in the world, a text-generating tool called GPT-3 that can write creative fiction, translate legalese into plain English, and answer obscure trivia questions. It’s the latest feat of intelligence achieved by deep learning, a machine learning method patterned after the way neurons in the brain process and store information. But it came at a hefty price: at least $4.6 million and 355 years in computing time, assuming the model was trained on a standard neural network chip, or GPU. The model’s colossal size — 1,000 times larger than a typical language model — is the main factor in its high cost. “You have to throw a lot more computation at something to get a little improvement in performance,” says Neil Thompson, an MIT researcher who has tracked deep learning’s unquenchable thirst for computing. “It’s unsustainable. We have to find more efficient ways to scale deep learning or develop other technologies.” Some of the Continue reading Shrinking deep learning’s carbon footprint→

The current generation of neural implants record enormous amounts of neural activity, then transmit these brain signals through wires to a computer. But, so far, when researchers have tried to create wireless brain-computer interfaces to do this, it took so much power to transmit the data that the implants generated too much heat to be safe for the patient. A new study suggests how to solve his problem — and thus cut the wires. More details

TikTok’s most valuable assets, a hyper-effective algorithm and a community of popular creators, may not be as easy to acquire as the company itself. More details

As part of the MIT Task Force on the Work of the Future’s new series of research briefs, Professor John Leonard teamed with professor of aeronautics and astronautics and of history David Mindell and with doctoral candidate Erik Stayton to explore the future of autonomous vehicles (AV) — an area that could arguably be called the touchstone for the discussion of jobs of the future in recent years. Leonard is the Samuel C. Collins Professor of Mechanical and Ocean Engineering in the Department of Mechanical Engineering, a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL), and member of the MIT Task Force on the Work of the Future. His research addresses navigation and mapping for autonomous mobile robots operating in challenging environments.  Their research brief, “Autonomous Vehicles, Mobility, and Employment Policy: The Roads Ahead,” looks at how the AV transition will affect jobs and explores how sustained investments in workforce training Continue reading 3 Questions: John Leonard on the future of autonomous vehicles→

Letter writers across the country will soon have a fun and beautiful new Forever stamp to choose from, featuring novel research from the Media Lab’s Biomechatronics research group.  The stamp is part of a new U.S. Postal Service (USPS) series on innovation, representing computing, biomedicine, genome sequencing, robotics, and solar technology. For the robotics category, the USPS chose the bionic prosthesis designed and built by Matt Carney PhD ’20 and members of the Biomechatronics group, led by Professor Hugh Herr. The image used in the stamp was taken by photographer Andy Ryan, whose portfolio spans images from around the world, and who for many years has been capturing the MIT experience — from stunning architectural shots to the research work of labs across campus. Ryan suggested the bionic work of the biomechatronics group to USPS to represent the future of robotics. Ryan also created the images that became the computing and solar Continue reading New US postage stamp highlights MIT research→

In recent years, entire industries have popped up that rely on the delicate interplay between human workers and automated software. Companies like Facebook work to keep hateful and violent content off their platforms using a combination of automated filtering and human moderators. In the medical field, researchers at MIT and elsewhere have used machine learning to help radiologists better detect different forms of cancer.  What can be tricky about these hybrid approaches is understanding when to rely on the expertise of people versus programs. This isn’t always merely a question of who does a task “better;” indeed, if a person has limited bandwidth, the system may have to be trained to minimize how often it asks for help. To tackle this complex issue, researchers from MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) have developed a machine learning system that can either make a prediction about a task, or defer the decision Continue reading An automated health care system that understands when to step in→