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.
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.”
MIT researchers have developed a new procedure that uses game theory to improve the accuracy and consistency of large language models (LLMs), reports Steve Nadis for Quanta Magazine. “The new work, which uses games to improve AI, stands in contrast to past approaches, which measured an AI program’s success via its mastery of games,” explains Nadis.
Researchers from MIT and elsewhere have used an AI model to develop a “new approach to finding money laundering on Bitcoin’s blockchain,” reports Andy Greenberg for Wired. “Rather than try to identify cryptocurrency wallets or clusters of addresses associated with criminal entities such as dark-web black markets, thieves, or scammers, the researchers collected patterns of bitcoin transactions that led from one of those known bad actors to a cryptocurrency exchange where dirty crypto might be cashed out,” explains Greenberg.
MIT researchers have developed a machine-learning accelerator chip to make health-monitoring apps more secure, reports Aman Tripathi for Interesting Engineering. “The researchers subjected this new chip to intensive testing, simulating real-world hacking attempts, and the results were impressive,” explains Tripathi. “Even after millions of attempts, they were unable to recover any private information. In contrast, stealing data from an unprotected chip took only a few thousand samples.”
MIT researchers have developed a new technique “that could allow most large language models (LLMs) like ChatGPT to retain memory and boost performance,” reports Tony Ho Tran for the Daily Beast. “The process is called StreamingLLM and it allows for chatbots to perform optimally even after a conversation goes on for more than 4 million words,” explains Tran.
MIT researchers have used machine learning to uncover the different kinds of sentences that most likely to activate the brain’s key language processing centers, reports Kyle Wiggers and Devin Coldewey for TechCrunch. The model, “was able to predict for novel sentences whether they would be taxing on human cognition or not,” they explain.
Researchers from MIT and elsewhere have developed a new AI technique for teaching robots to pack items into a limited space while adhering to a range of constraints, reports Nick Hilden for Scientific American. “We want to have a learning-based method to solve constraints quickly because learning-based [AI] will solve faster, compared to traditional methods,” says graduate student Zhutian “Skye” Yang.
Researchers from MIT and Harvard have explored astrocytes, a group of brain cells, from a computational perspective and developed a mathematical model that shows how they can be used to build a biological transformer, reports Kyle Wiggers for TechCrunch. “The brain is far superior to even the best artificial neural networks that we have developed, but we don’t really know exactly how the brain works,” says research staff member Dmitry Krotov. “There is scientific value in thinking about connections between biological hardware and large-scale artificial intelligence networks. This is neuroscience for AI and AI for neuroscience.
MIT researchers have developed SoftZoo, “an open framework platform that simulated a variety of 3D model animals performing specific tasks in multiple environmental settings,” reports Andrew Paul for Popular Science. “This computational approach to co-designing the soft robot bodies and their brains (that is, their controllers) opens the door to rapidly creating customized machines that are designed for a specific task,” says CSAIL director, Prof. Daniela Rus.
Researchers at MIT have developed “SoftZoo,” a platform designed to “study the physics, look and locomotion and other aspects of different soft robot models,” reports Brian Heater for TechCrunch. “Dragonflies can perform very agile maneuvers that other flying creatures cannot complete because they have special structures on their wings that change their center of mass when they fly,” says graduate student Tsun-Hsuan Wang. “Our platform optimizes locomotion the same way a dragonfly is naturally more adept at working through its surroundings.”
Researchers at MIT have created a four-legged robot called DribbleBot, reports Caroline Goggin for WHDH. The robot “can dribble a soccer ball under the same conditions as humans, using onboard sensors to travel across different types of terrain.”
Popular Science reporter Andrew Paul spotlights how researchers from MIT CSAIL have developed a soccer-playing robot, dubbed DribbleBot, that can handle a variety of real-world terrains. “DribbleBot showcases extremely impressive strides in articulation and real-time environmental analysis. Using a combination of onboarding computing and sensing, the team’s four-legged athlete can reportedly handle gravel, grass, sand, snow, and pavement, as well as pick itself up if it falls.”
MIT researchers have created “Dribblebot,” a four-legged robot capable of playing soccer across varying terrain, reports Brian Heater for TechCrunch.
Researchers at MIT have created a four-legged robot capable of dribbling a soccer ball and running across a variety of terrains, reports Ross Cristantiello for Boston.com. “Researchers hope that they will be able to teach the robot how to lift a ball over a step in the future,” writes Cristantiello. “They will also explore how the technology behind DribbleBot can be applied to other robots, allowing machines to quickly transport a range of objects around outside using legs and arms.”