The research is being led in collaboration with TILOS Robotics team member Dr. Nikolay Atanasov, head of the Existential Robotics Laboratory at the UC San Diego Jacobs School of Engineering, whose work focuses on robotic perception, semantic mapping, and autonomous systems.
“Simultaneous localization and mapping, or SLAM, helped move robots beyond fixed industrial settings and into more dynamic environments,” said Dr. Nikolay Atanasov, TILOS Robotics team member and faculty member in the Jacobs School’s Department of Electrical and Computer Engineering at UC San Diego.
TILOS Chips team co-lead Farinaz Koushanfar is one of eight UC San Diego faculty elected 2025 Fellows of the American Association for the Advancement of Science (AAAS). Congratulations to all!
TILOS Chips and Foundations team member Sicun (Sean) Gao was honored with a 2024-2025 UC San Diego Jacobs School of Engineering Best Undergraduate Teacher Award.
TILOS Director Yusu Wang, Deputy Director Arya Mazumdar, and Foundations team member Mikhail Belkin are the newest Endowed Chairs in AI and Data Science at UC San Diego’s Halıcıoğlu Data Science Institute, the administrative home of TILOS.
A team led by Adityanarayanan Radhakrishnan at the Massachusetts Institute of Technology and Mikhail Belkin at the University of California San Diego has developed a method to hunt down and manipulate hidden concepts lurking inside large language models — the AI systems behind ChatGPT, Claude and their kin. Their approach, published this week in Science, can […]
In a pioneering breakthrough for artificial intelligence research, a team of scientists has unveiled a novel technique to precisely steer the output of large language models (LLMs) by manipulating specific internal concepts encoded within these models. This innovative approach promises significant advancements in making LLMs more reliable, efficient, and adaptable, while simultaneously shedding light on […]
A team of researchers has found a way to steer the output of large language models by manipulating specific concepts inside these models. The new method could lead to more reliable, more efficient, and less computationally expensive training of LLMs. But it also exposes potential vulnerabilities.
