Download RSS feed: News Articles / In the Media / Audio
Researchers at MIT have developed a new model for training robots dubbed Heterogeneous Pretrained Transformers (HPT), reports Brain Heater for TechCrunch. The new model “pulls together information from different sensors and different environments,” explains Heater. “A transformer was then used to pull together the data into training models. The larger the transformer, the better the output. Users then input the robot design, configuration, and the job they want done.”
MIT researchers have developed a new training technique called Heterogeneous Pretrained Transformers (HPT) that could help make general-purpose robots more efficient and adaptable, reports Christopher Isak for TechAcute. “The main advantage of this technique is its ability to integrate data from different sources into a unified system,” explains Isak. “This approach is similar to how large language models are trained, showing proficiency across many tasks due to their extensive and varied training data. HPT enables robots to learn from a wide range of experiences and environments.”
Liquid AI, an MIT startup, is unveiling a new AI model based on a liquid neural network that “has the potential to be more efficient, less power-hungry, and more transparent than the ones that underpin everything from chatbots to image generators to facial recognition systems, reports Will Knight for Wired.
Graduate student Nouran Soliman speaks with NBC Boston about the use of “personhood credentials,” a new technique that can be used to verify online users as human beings to help combat issues such as fraud and misinformation. “We are trying to also think about ways of implementing a system that incorporates personal credentials in a decentralized way,” explains Soliman. “It's also important not to have the power in one place because that compromises democracy.”
Researchers at MIT and elsewhere have developed a 3D printing method that allows “precise control over color, shade, texture, all with just a single material,” reports Vanesa Listek for 3Dprint.com. This technique “promises a faster and more sustainable solution than traditional approaches relying on multiple materials and nozzle changes,” explains Listek.
Researchers at MIT and elsewhere have developed “a new method of 3D printing that uses heat-responsive materials to print multi-color and multi-textured objects in one step,” reports Laura Griffiths for TCT Magazine. “The method has so far been tested using three heat-responsive filaments including a foaming polymer with particles that expand as they are heated, and wood and cork fiber-filled filaments,” explains Griffiths.
Researchers at MIT have developed GenSQL, a new generative AI system that can be used “to ease answering data science questions,” reports Allison Proffitt for Bio-It World. “Look how much better data science could be if it was easier to use,” says Research Scientist Mathieu Huot. “It’s not perfect yet, but we believe it’s quite an improvement over other options.”
Researchers at MIT have found large language models “often struggle to handle more complex problems that require true understanding,” reports Kirimgeray Kirimli for Forbes. “This underscores the need for future versions of LLMs to go beyond just these basic, shared capabilities,” writes Kirimli.
Prof. Regina Barzilay has received the WebMD Health Heros award for her work developing a new system that uses AI to detect breast cancer up to 5 years earlier, reports WHDH. “We do have a right to know our risk and then we, together with our healthcare providers, need to manage them,” says Barzilay.
Writing for The Washington Post, Prof. Daniela Rus, director of CSAIL, and Nico Enriquez, a graduate student at Stanford, make the case that the United States should not only be building more efficient AI software and better computer chips, but also creating “interstate-type corridors to transmit sufficient, reliable power to our data centers.” They emphasize: “The United States has the talent, investor base, corporations and research institutions to write the most advanced AI models. But without a powerful data highway system, our great technology advances will be confined to back roads.”
Researchers from MIT and elsewhere have created an AI Risk Repository, a free retrospective analysis detailing over 750 risks associated with AI, reports Tor Constantino for Forbes. “If current understanding is fragmented, policymakers, researchers, and industry leaders may believe they have a relatively complete shared understanding of AI risks when they actually don’t,” says Peter Slattery, a research affiliate at the MIT FutureTech project. “This sort of misconception could lead to critical oversights, inefficient use of resources, and incomplete risk mitigation strategies, which leave us more vulnerable.”
In an article for Forbes, Robert Clark spotlights how MIT researchers developed a new model to predict irrational behaviors in humans and AI agents in suboptimal conditions. “The goal of the study was to better understand human behavior to improve collaboration with AI,” Clark writes.
Researchers at MIT have developed “a publicly available database, culled from reports, journals, and other documents to shed light on the risks AI experts are disclosing through paper, reports, and other documents,” reports Jon McKendrick for Forbes. “These benchmarked risks will help develop a greater understanding the risks versus rewards of this new force entering the business landscape,” writes McKendrick.
A new database of AI risks has been developed by MIT researchers in an effort to help guide organizations as they begin using AI technologies, reports Will Knight for Wired. “Many organizations are still pretty early in that process of adopting AI,” meaning they need guidance on the possible perils, says Research Scientist Neil Thompson, director of the FutureTech project.
MIT researchers have developed an AI risk repository that includes over 70 AI risks, reports Kyle Wiggers for TechCrunch. “This is an attempt to rigorously curate and analyze AI risks into a publicly accessible, comprehensive, extensible and categorized risk database that anyone can copy and use, and that will be kept up to date over time,” explains Peter Slattery, a research affiliate at the MIT FutureTech project.