weekend ai reads for 2023-10-06

📰 ABOVE THE FOLD: USER INTERFACING

• Multi-Modal AI is a UX Problem Matt Rickard

  • poses a thoughtful question

• via phillip, I was curious if I could practice my Russian with ChatGPT-audio. Yup. Speakers move between languages effortlessly. (2:17)   dmvaldman, Twitter (sorry)

  • amazing demo

• Text Is the Universal Interface Scale.ai

• Ubiquitous cameras plus AI will change how we understand the world Understanding AI

• Osmo — Giving computers a sense of smell.

• Oracle brings voice-activated AI to healthcare with Clinical Digital Assistant Venture Beat

📻 QUOTES OF THE WEEK

(source)

Jesper Hansen, composer (source)

🏗️ FOUNDATIONS & CULTURE

• Bullet Points: Oh-just-shut-up edition The Future, Now and Then, Substack

  • first third is a cogent critique of Ethan Mollick’s “jagged frontier”

• What Happens When Kareem Attends first AI Conference Kareem Abdul-Jabbar, Substack

• Third-party AI tools pose increasing risks for organizations MIT Sloan School of Management

  • related, Board Practices: Artificial intelligence Harvard Law School Forum on Corporate Governance

• Tom Hanks Warns Fans About ‘AI Version of Me’ Promoting Dental Plan: ‘I Have Nothing to Do With It’ Variety

  • related, MrBeast warns followers about deepfake scam — The reigning YouTube creator called out AI-generated imagery as a “serious problem”.   Mashable

• Wanted: High Performers for the Last Job You’ll Ever Have — Train an AI to replace yourself. Salary paid and profits shared indefinitely.   Every

🎓 EDUCATION

• Using AI While Protecting Student Data AutomatED: Teaching Better with Tech, Beehiiv

• So Much for ‘Learn to Code’ — In the age of AI, computer science is no longer the safe major.   The Atlantic ($)

• We now have a pipeline that goes: MIT OCW -> Syllabus -> Table of Contents -> Textbook. Owen Colegrove, Twitter (sorry)

  • Introduction to Deep Learning (AI-generated textbook)

• Fewer essays, more oral exams, no online tests: shake-up to prevent students using AI to cheat Irish Independent

• How to Promote Responsible Open Foundation Models — Experts from industry, academia, and government share lessons learned and outline a path forward at a Princeton-Stanford workshop.   Stanford University Human-Centered Artificial Intelligence

  • summary at the link above; full workshop here (6:03:15)

• Artificial intelligence such as ChatGPT to be allowed in Australian schools from 2024 The Guardian

📊 DATA & TECHNOLOGY

• Llamar.ai: A deep dive into the (in)feasibility of RAG with LLMs Bold Data

  • related (1), via adam, Why you should build RAG from scratch - with Jerry Liu from LlamaIndex (1:08:05)   Latent Space 

  • related (2), When to choose RAG vs Finetuning? [JPG]   source unknown, Twitter (sorry)

• via adam, Tiny Language Models Come of Age Quanta Magazine

• Computational Power and AI AI Now Institute

• Vizro McKinsey, Github

  • demonstration site

• Mistral 7B — The best 7B model to date, Apache 2.0

  • this model apparently was not shaped for safety:

• Can AI Beat the Market? Wall Street Is Desperate to Try — Those seeking to build the perfect money machine are discovering the technology doesn’t always work as planned.   Bloomberg

🎉 FUN and/or PRACTICAL THINGS

• Weird A.I. Yankovic, a cursed deep dive into the world of voice cloning Waxy (Andy Baio)

  • “monstrous” is right

  • one example: Weird Al singing “Beat It” and Michael Jackson singing “Eat It”

• RealFill — Reference-Driven Generation for Authentic Image Completion   Realfill, Github

  • completes images based on reference photos; seemingly does 3d mapping of reference images to “fill” outside borders of the target image; will make more sense once you see it

• How to Spot AI Generated Images Everypixel Journal

• via adam, Translating Latin demonology manuals with GPT-4 and Claude Res Obscura, Substack

• ‘You’ve got to be data-driven’: the fashion forecasters using AI to predict the next trend The Guardian

• found out that facebook messenger has ai generated stickers now and I don't think anyone involved has thought anything through Pioldes, Twitter (sorry)

  • “Waluigi rifle”, “Trudeau buttocks”, and other beauts

🧿 AI-ADJACENT

• Bill Gurley presents 2,851 Miles (36:03)   All-In Podcast, YouTube