Talks
A Visual Exploration Under the Hood of a Large Language Model
Slides
Speaker: Chris Biow
Start time: 6:15pm
It can seem easier to get answers from prompt-based interfaces like ChatGPT than from actual humans. However a responsible technologist should understand layers of abstraction behind the prompt or API that outputs results. These layers determine the behavior and risks at every level, down to the logic gates we are invoking.
We will explore how an LLM works at the mathematical layer of abstraction. (Only addition and multiplication are required!) LLM practitioners from Python coders to interactive prompt users will better understand the innovations behind the remarkable AI advances of the last few years and what may be coming.
Using open source 3D visualization models, we will follow the entire inferencing process in an LLM, stepping through each mathematical operation: input and position embedding, the transformer (including layer normalization, self-attention, projection, and feed-forward multi-layer perceptron), and the final normalization and output.
Disrupting Venture Capital With AI: A New Era for Startups and Investors
Speaker: Sheamus McGovern
Start time: 7:15pm
The integration of AI into venture capital signifies a paradigm shift for both startups and investors. For startups, it means increased opportunities to be discovered and funded based on merit and potential, analyzed through data. For investors, AI offers tools to mitigate risks and identify high-potential investments with greater efficiency, transforming the traditional venture capital landscape into a more dynamic, informed, and competitive arena. In this presentation, we will delve into some of the technical details and AI tools for venture capital we’ve developed using large language models (LLMs), highlighting how we leverage data sources and trends across various datasets. This will illustrate the practical application and impact of our AI-driven solutions in the VC ecosystem.
Chris Biow has over 30 years background as a technologist working in text retrieval, databases, and machine learning. He holds a BS in mathematics from the US Naval Academy and an MS in computer science from the University of Maryland. He served with the US Navy as an F-14 Tomcat RIO (Radar Intercept Officer). Upon leaving active duty with the Navy, Chris founded a sales-enablement software company and then worked delivering public sector and commercial solutions with search and database software at Verity, Autonomy, as Federal CTO at MarkLogic, and at MongoDB. He specialized in problematically large textual data problems and was the first to stand up a petabyte database on MongoDB. He has been a speaker at AWS Re:Invent, NoSQL Now, Text Analytics, and Smart Data conferences. He most recently led BasisTech’s Rosette AI-based text analytics software through a period of fast growth, leading to an acquisition by Veritas Capital in December 2022 and a merger with Babel Street Ltd. He is now back with BasisTech as an Entrepreneur in Residence.
As a technologist, Chris has always believed that those applying technologies to business problems should strive to maintain an understanding of their technologies down to the level of the mathematical operation, machine register state, and cache chain.
Sheamus McGovern is the founder of ODSC (The Open Data Science Conference), hosting major AI conferences in cities like Boston, San Francisco, and London. He also serves as a Venture Partner and Head of AI at Cortical Ventures, an investment fund with a primary focus on AI. As a software architect, data engineer, and AI enthusiast, Sheamus began his career in finance, working for various financial institutions and quantitative hedge funds. Over the past decade, he has consulted with numerous companies and startups, developing cutting-edge, data-driven applications in sectors such as finance, healthcare, and eCommerce.His academic credentials include degrees from Northeastern University, Boston University, Harvard University, and a Certificate in Quantitative Finance (CQF).
