Personalization features can make LLMs more agreeable
The context of long-term conversations can cause an LLM to begin mirroring the user’s viewpoints, possibly reducing accuracy or creating a virtual echo-chamber.
Enabling small language models to solve complex reasoning tasks
The “self-steering” DisCIPL system directs small models to work together on tasks with constraints, like itinerary planning and budgeting.
Researchers discover a shortcoming that makes LLMs less reliable
Large language models can learn to mistakenly link certain sentence patterns with specific topics — and may then repeat these patterns instead of reasoning.
Teaching large language models how to absorb new knowledge
With a new method developed at MIT, an LLM behaves more like a student, writing notes that it studies to memorize new information.
MIT researchers propose a new model for legible, modular software
The coding framework uses modular concepts and simple synchronization rules to make software clearer, safer, and easier for LLMs to generate.
MIT-IBM Watson AI Lab researchers have developed a universal guide for estimating how large language models will perform based on smaller models in the same family.
Language models follow changing situations using clever arithmetic, instead of sequential tracking. By controlling when these approaches are used, engineers could improve the systems’ capabilities.
Researchers find nonclinical information in patient messages — like typos, extra white space, and colorful language — reduces the accuracy of an AI model.
Unpacking the bias of large language models
In a new study, researchers discover the root cause of a type of bias in LLMs, paving the way for more accurate and reliable AI systems.
Words like “no” and “not” can cause this popular class of AI models to fail unexpectedly in high-stakes settings, such as medical diagnosis.