Beyond the Final Frontier: Crafting Star Trek with Language Models
Large language models have come in to vogue with the advent of ChatGPT. In 2023 the Writers Guild of America went on strike to protect their members from being forced to compete with LLMs writing scripts for shows. But some of us have been using LLMs to create new television scripts for almost a decade! In this talk we’ll follow the speaker’s journey to create new quality Star Trek content, starting with Neural Machine Translation (NMT) in 2015, and subsequent attempts through the late 2010s and up through present. This talk will be a narrative history of LLMs and techniques developed over the years for hierarchically creating data.
A passing familiarity with ChatGPT will be a suitable technical background for this talk, as the speaker guides you on a historical journey on the hype cycles and pit falls he’s fallen for over the years to develop new content. Spanning from NMT, RNNs, LSTMs to modern LLMs such as GPT2 (not so modern now), and GPT 3, 3.5, and 4. He will also showcase digital artifacts from Twitter and a website he created.
The talk will be a treat for anyone who (1) likes Star Trek or (2) likes to be obnoxious with their knowledge of obscure historical trivia at parties.