Prof. Pulkit Agrawal speaks with NPR Short Wave host Regina Barber and science correspondent Geoff Brumfiel about his work developing a new technique that allows robots to train in simulations of scanned home environments. “The power of simulation is that we can collect very large amounts of data,” explains Agrawal. “For example, in three hours' worth of simulation, we can collect 100 days' worth of data.”
Researchers at MIT have developed “Clio,” a new technique that “enables robots to make intuitive, task-relevant decisions,” reports Jennifer Kite-Powell for Forbes. The team’s new approach allows “a robot to quickly map a scene and identify the items they need to complete a given set of tasks,” writes Kite-Powell.
Researchers at MIT have developed a new method that “enables robots to intuitively identify relevant areas of a scene based on specific tasks,” reports Baba Tamim for Interesting Engineering. “The tech adopts a distinctive strategy to make robots effective and efficient at sorting a cluttered environment, such as finding a specific brand of mustard on a messy kitchen counter,” explains Tamim.
Writing for Fast Company, Moshe Tanach highlights how researchers from the MIT Lincoln Laboratory Supercomputing Center are developing new technologies to reduce AI energy costs, such as power-capping hardware and tools that can halt AI training.
MIT researchers have improved upon the diffusion models used in AI image generation, reports Alok Jha for The Economist. Working with electrically charged particles, the team created “Poisson flow generative models,” which “generate images of equal or better quality than state-of-the-art diffusion models, while being less error-prone and requiring between ten and 20 times fewer computational steps,” Jha explains.
Researchers at CSAIL have created three “libraries of abstraction” – a collection of abstractions within natural language that highlight the importance of everyday words in providing context and better reasoning for large language models, reports Darren Orf for Popular Mechanics. “The researchers focused on household tasks and command-based video games, and developed a language model that proposes abstractions from a dataset,” explains Orf. “When implemented with existing LLM platforms, such as GPT-4, AI actions like ‘placing chilled wine in a cabinet' or ‘craft a bed’ (in the Minecraft sense) saw a big increase in task accuracy at 59 to 89 percent, respectively.”
In a new study examining the potential impact of AI on jobs that employ computer vision, MIT researchers found, “a vast majority of jobs previously identified as vulnerable to AI are not economically beneficial for employers to automate at this time,” reports Catherine Thorbecke for CNN. “In many cases, humans are the more cost-effective way, and a more economically attractive way, to do work right now,” says Research Scientist Neil Thompson, director of the FutureTech Research Project at CSAIL. “What we’re seeing is that while there is a lot of potential for AI to replace tasks, it’s not going to happen immediately.”
A new working paper by MIT researchers finds that artificial intelligence is not currently a cost-effective replacement in jobs where computer vision is employed, reports Saritha Rai for Bloomberg. “Our study examines the usage of computer vision across the economy, examining its applicability to each occupation across nearly every industry and sector,” explains Research Scientist Neil Thompson, director of the FutureTech Research Project at CSAIL. “We show that there will be more automation in retail and healthcare, and less in areas like construction, mining or real estate.”
A new working paper by MIT researchers focuses on whether human work, including vision tasks, are worth replacing with AI computer vision, reports Jeremy Hsu for New Scientist. “There are lots of tasks that you can imagine AI applying to, but actually cost-wise you just wouldn’t want to do it,” says Research Scientist Neil Thompson, director of the FutureTech Research Project at CSAIL.
Researchers at MIT have released a new working paper that aims to quantify the severity and speed with which AI systems could replace human workers, reports Hiawatha Bray for The Boston Globe. The paper concluded that “it’s not enough for AI systems to be good at tasks not performed by people,” explains Bray. “The system must be good enough to justify the cost of installing it and redesigning the way a job is done.”
Javier Ramos '12, SM '14, co-founder of InkBit, and his colleagues have developed a, “3D inkjet printer that uses contact-free computer vision feedback to print hybrid objects with a broad range of new functional chemistries,” reports Ed Brown for Tech Briefs. “Our vision for Inkbit is to reshape how the world thinks about production, from design to execution and make our technology readily available,” says Ramos. “The big opportunity with 3D printing is how to disrupt the world of manufacturing — that’s what we're focused on.”
Researchers from MIT and elsewhere have developed a new 3D printing process that “allows users to create more elastic materials along with rigid ones using slow-curing polymers,” reports Tony Ho Tran for the Daily Beast. The researchers used the system to create a, “3D printed hand complete with bones, ligaments, and tendons. The new process also utilizes a laser sensor array developed by researchers at MIT that allows the printer to actually ‘see’ what it’s creating as it creates it.”
MIT researchers have developed a new tool called “PhotoGuard” that can help protect images from AI manipulation, reports Ross Cristantiello for Boston.com. The tool “is designed to make real images resistant to advanced models that can generate new images, such as DALL-E and Midjourney,” writes Cristantiello.
Researchers at MIT have developed “PhotoGuard,” a tool that can be used to protect images from AI manipulation, reports Catherine Thorbecke for CNN. The tool “puts an invisible ‘immunization’ over images that stops AI models from being able to manipulate the picture,” writes Thorbecke.
MIT researchers have developed a new machine-learning technique that can identify which pixels in an image represent the same material, which could help with robotic scene understanding, reports Kyle Wiggers for TechCrunch. “Since an object can be multiple materials as well as colors and other visual aspects, this is a pretty subtle distinction but also an intuitive one,” writes Wiggers.