Talk Series for the San Francisco Public Library
By Chris Mentzel (all opinions his own)
📅 Latest talk: September 9, 2025 — SFPL Main Branch
Chris Mentzel, Executive Director of Stanford Data Science, outlines the fascinating journey of Artificial Intelligence (AI) from its early beginnings to its role in shaping modern life. This talk covers the history, current applications, and future potential of AI, highlighting how it touches many aspects of daily living.
Duration: 45-60 minutes including Q&A
Audience: No prerequisites - suitable for all audiences
-
The Creativity Code — Marcus du Sautoy
Explores how AI is learning to write, paint, and think creatively. ISBN: 978-0674244713 -
Weapons of Math Destruction — Cathy O'Neil
Examines how big data increases inequality and threatens democracy.
SFPL: Available in collection -
Human Compatible — Stuart Russell
Addresses AI safety and the challenge of controlling artificial intelligence.
SFPL: Available in collection -
Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence — Erik Brynjolfsson, Bharat Chandar, Ruyu Chen (Aug 26, 2025)
Read online
-
AI Video (OpenAI Sora) — Prompt:
“A green tiger wearing a scarf is learning through a jungle, carrying a cup, filmed in the style of film noir.”
Watch Video -
AI Image (ChatGPT 5) — Prompt:
“A sweet gorilla wearing a bow tie is standing in front of Vermont, holding a camera, painted in the style of deco noir.”
View Image -
NotebookLM — Google’s notebook language model that lets users add documents, ask questions, and even generate podcasts.
Example Notebook
Working on cleaning the presentation1 for public sharing — no promises.
- Large Language Models (LLMs): The engines behind ChatGPT and friends. They’ve read a ton of text and can spin out human-like answers by predicting what comes next.
- Generative Adversarial Networks (GANs): Think of two AIs playing cat-and-mouse — one makes fakes, the other tries to catch them. Over time, the fakes get so good they fool us too.
- Synthetic Data: Made-up data that looks real. Super handy when you don’t have enough examples or can’t use the real thing for privacy reasons.
- Neural Networks: Loosely inspired by brains. Layers of little “nodes” passing signals around until patterns pop out.
- Transformers: The special recipe that made modern AI take off. They let models look at whole chunks of information at once instead of word by word.
- RAG (Retrieval-Augmented Generation): Fancy name for “open-book AI.” The model pulls from your docs or a database so it’s not just making things up from memory.
- Reinforcement Learning (RL): Trial-and-error learning. The AI tries things, gets rewards or penalties, and improves over time. Think of training a puppy, but digital.
- RLHF (Reinforcement Learning with Human Feedback): A twist on RL where humans give the feedback. People rank answers, and the AI learns which ones we like.
- Hallucination: When AI sounds confident but is flat-out wrong.
- Prompt Engineering: Talking to AI in just the right way to get what you want. Feels a bit like learning a new language.
- Fine-tuning: Taking a general-purpose model and teaching it a new specialty, like giving a chef lessons in sushi-making.
- Tokenization: Breaking text into bite-sized chunks (words or pieces of words) so computers can handle them.
- Embeddings: Turning words or ideas into math-y coordinates so the AI can “feel” what’s close in meaning.
- Agentic AI: Not just answering questions, but actually doing things — booking your trip, organizing your files, running small projects.
- Computer Vision: Your phone unlocking with Face ID, or a car spotting a stop sign. That’s AI “seeing.”
- Natural Language Processing (NLP): Basically, AI understanding and generating language — from translating Spanish menus to writing your emails.
- Multimodal AI: A model that can juggle text, images, audio, maybe even video all at once.
- Edge AI: AI running right on your phone or camera, so it’s faster and more private — no cloud required.
- AI Alignment: Teaching AI to follow human values — in spirit, not just letter of the law.
- Model Collapse: When AIs keep learning from AI-written stuff, and the quality slowly drifts downhill.
- Bias in AI: Garbage in, garbage out — if the data’s biased, the AI will be too.
- Explainable AI (XAI): Peeking under the hood so we know why the AI said what it said.
- Constitutional AI: Giving the AI a moral compass baked right into its training.
Footnotes
-
The presentation was created using AI tools, including ChatGPT, Claude, and Gemini for outline, fact-checking, references, and demo suggestions. Gamma.app was used for the initial PowerPoint layout and image creation. ↩