Caterina Constantinescu - Deconstructing LLM Use: Key Considerations to Deliver Custom Solutions
Deconstructing LLM Use: Key Considerations to Deliver Custom Solutions by Caterina Constantinescu
Visit https://rstats.ai/nyr to learn more.
Abstract: The rate of advancement in AI research (and LLMs specifically) won't have escaped anybody's attention by now. This degree of progress naturally lends itself to hot takes, memes and dramatisation, when reality is much more nuanced. This talk will explore why AI won't 'take away our jobs' just yet, by discussing some of the real-world constraints and customisations still required. For instance, licensing, privacy and data ownership are legitimate talking points before large-scale adoption in industry. In addition, data collection/scraping for model fine-tuning might also be non-trivial to implement depending on the specific scenario at hand, and experience/UI design may similarly require considerable conscious thought before LLM deployment / integration. In this talk, I will discuss these issues (and more!) to highlight that LLM use involves a high degree of subtlety and customisation, acting as counterweights to now common hyperbolae on the topic.
Bio: I am a data scientist and enthusiast with an academic background in psychological research methods. I've subsequently worked across diverse areas such as national health data, occupational therapy, transport, data for good, tourism, finance and more recently, ESG - applying a mix of data design, analysis and visualisation across all of these. For several years I have also organised meetings for the R user group in Edinburgh. Currently, I work as Principal Data Consultant at GlobalLogic, collaborating across multiple teams to design technical solutions and provide steering on strategic roadmap development.
Twitter: / c__constantine
Presented at the 2023 New York R Conference (July 14, 2023)