weekend ai reads for 2023-07-07

📰 ABOVE THE FOLD: AUTONOMY

• LLM Powered Autonomous Agents Lilian Weng

  • Comprehensive overview of necessary components

• Inside the race to build an ‘operating system’ for generative AI Venture Beat

• Self-Improving Robots: Embracing Autonomy in Robot Learning The Stanford AI Lab Blog

• RoboCat: A self-improving robotic agent Google Deep Mind

• What AI can do with a toolbox... Getting started with Code Interpreter — Democratizing data analysis with AI One Useful Thing (aka: Ethan Mollick), Substack

  • Available to ChatGPT+ subscribers “next week”

📻 QUOTE OF THE WEEK

R. Scott SD (Twitter, sorry)

🏗️ FOUNDATIONS & CULTURE

• Will AI Change Our Memories? (6:41) Nerdwriter, YouTube

• AI moderation is no match for hate speech in Ethiopian languages — New research raises concerns that moderation tools can’t detect violent threats in Tigray and elsewhere. Rest of World

• EU launches four new testing facilities to develop responsible AI The Next Web

  • The facilities are Manufacturing, Healthcare, Agriculture & food, and Cities & communities

• Will AI interfere with relationships? Gen Z thinks so. Ideo

• Reid Hoffman on the Possibilities of AI (Ep. 183) [transcript] — From creating a thousand games to talking to dolphins, Reid is pumped about what AI will allow him to do. Conversations with Tyler

• Grocers plan to increase AI spending by 400% before 2025 Supermarket News

  • Possibly relevant to other retailers as well

🎓 EDUCATION

• Generative AI for Programming Education: Benchmarking ChatGPT, GPT-4, and Human Tutors arXiv

• How elite schools like Stanford became fixated on the AI apocalypse — A billionaire-backed movement is recruiting college students to fight killer AI, which some see as the next Manhattan Project. Washington Post

• via andy, Emerging Technology and AI [PDF, 29 page report, n=213]  Inside Higher Ed

• ChatGPT Drops About 10% in Traffic as the Novelty Wears Off Similar Web

  • One hypothesis is the drop is linked to students no longer using ChatGPT for coursework. It will be interesting to see how seasonal the usage is.

📊 DATA & TECHNOLOGY

• Massive Leak Of ChatGPT Credentials: Over 100,000 Accounts Affected Search Engine Journal

• Can you trust ChatGPT’s package recommendations? — ChatGPT can offer coding solutions, but its tendency for hallucination presents attackers with an opportunity. Here's what we learned. Vulcan

• Unraveling GPU Inference Costs for Fine-tuned Open-source Models V/S Closed Platforms MLOps Community

• Models generating training data: huge win or fake win? Dave Summarizes Papers, Substack

🎉 FUN and/or PRACTICAL THINGS

• Arlo - AI alarm clock iOS only

  • AI-simulated celebrity voices (Morgan Freeman! Oprah! Kobe?), and GPT-generated summaries of your day as your alarm. Free tier works okay. And doesn’t collect data.

  • related (?): How actors are losing their voices to AI — Performers forced to compete with themselves as companies’ use of technology for cloning prompts calls to update copyright law Financial Times

• Whose generated line is it anyway? AI tries to crack humour’s DNA The Guardian

• Machine learning helps researchers identify hit songs with 97% accuracy — Scientists showed that applying machine learning to neural data leads to almost perfect classification accuracy for songs that may become hits American Association for the Advancement of Science

• Welcome to your AI YC mentor — YC mentor allows you to get knowledge from the best accelerator in the world. Just ask a question to get started.

• Jeffrey Celavie - AI Oracle, AI Astrologer

  • DO NOT SHARE PERSONAL INFORMATION ONLINE

  • Requests PII that may be linked to your identity. Site works fine if you make up a birth place, date, and time.

• Evita The world’s first AI companion for performers

  • Among its features are a voice coach and a Shakespeare translator. Currently requires sign-up for a waitlist.

🧿 AI-ADJACENT

• Phase transition in Random Circuit Sampling arXiv

  • Our knowledge of quantum computing comes from one meeting with IBM many years ago, and from as much as we understand of what Scott Aaronson shares on his blog. The consensus last summer seemed to be that QS would require 1,000s of qubits. Google claims their latest RCS experiment has an estimated computational cost of simulation of 47.2 years, which maybe is QS-adjacent? QS or not, this is interesting for the scientific advancement, and uninteresting because there still is no real-world utility demonstrated by these experiments, e.g., quantum machine learning.