WHAT: "Beyond Autocomplete: Instruction Following & Chain-of-Thought Reasoning in LLM Agents"
LEADER: Abraham Sanders
WHERE: Carnegie 113
EVENT PAGE: https://tw.rpi.edu/media/foci-llm-users-group-beyond-autocomplete-instruction-following-chain-thought-reasoning-llm
WHEN: 6p, 15 Nov (pizza et.al. at 5:30p)
CONTACT: Aaron Green <email@example.com>
DESCRIPTION: GPT-like LLMs are built on the principles of next-word prediction. However, the secret sauce that makes ChatGPT, Llama, Falcon & friends so useful is their ability to go beyond simple autocomplete tasks and model higher-order processes such as following instructions, reasoning, and planning. In this talk we take a look at how this is accomplished, and how LLMs can power tool-using dialogue agents that automate a wide variety of knowledge-bound tasks.
BIO: Abraham is a third-year PhD student in Cognitive Science at RPI, working in the LACAI lab with Dr. Tomek Strzalkowski. His research interests include open-domain and goal-oriented conversational agents along with multimodal, natural spoken dialogue systems. Previously, he was a lead software engineer at Nextech Systems where he worked on electronic health record system interoperability.
Slides and a recording of this talk are now available...
Recordings of previous FOCI Users Group sessions: