SBTB 2023: Leah McGuire, Keynote: LLMs, what are they good for?

ai.bythebay.io Nov 2025, Oakland, full-stack AI conference Prompt: In the style of Terry Pratchett write an abstract for a tech conference about how to decide on good use cases for and evaluate the performance of large language models. LLaMA2-70B: Greetings, fellow sentients! Are you tired of feeling like a mere mortal, limited by the constraints of your puny human brain? Fear not, for I have come to impart upon you the divine wisdom of the ages: how to decide on good use cases for and evaluate the performance of large language models. Firstly, let us consider the humble origins of these magnificent beasts. Born from the fiery depths of the digital underworld, these models have been fed on a feast of data, their hunger for knowledge insatiable. [Truncated to leave space for human generated content] In this talk, we will take a gander at those fancy, shiny new things that have become oh so trendy - large language models (LLMs). Dusting off my neuroscience PhD hat, we will get a little bit philosophical about intelligence, computational complexity, and how language fits into human cognition. Spoiler alert, the conclusion will not be that LLMs are sentient. However, they are both a powerful multitool and an incredibly compelling interface for humans. We will then take a look at some use cases for LLMs. We will take a peek at some lesions from the more mature, practically ancient history, explosion of generative art and how various use cases have panned out. We will then play a little game of LLM telephone and try to quantify what happens to the information we want to convey.…