weekend ai reads for 2023-10-20

📰 ABOVE THE FOLD: MULTIMODALITY

• Multimodality and Large Multimodal Models (LMMs) Chip Huyen

• Will AI kill our creativity? It could – if we don’t start to value and protect the traits that make us human The Conversation

• Reka launches Yasa-1, a multimodal AI assistant to take on ChatGPT Venture Beat

• LLark: A Multimodal Foundation Model for Music Spotify Research

  • the paper arXiv

• Multi-modal prompt injection image attacks against GPT-4V Simon Willison’s Blog

• How ‘A.I. Agents’ That Roam the Internet Could One Day Replace Workers The New York Times

  • related, AI Could Spur an Economic Boom. Humans Are in the Way. Wall Street Journal

📻 QUOTE OF THE WEEK

Grin Lord, clinical psychologist and CEO of mpathic.ai (source)

🏗️ FOUNDATIONS & CULTURE

• Borges and AI arXiv

  • related, via adam, Unbundling AI Benedict Evans

• via adam, State of AI Report Nathan Benaich, Air Street Capital

  • author’s summary

• Why AI must be decolonized to fulfill its true potential — Data gaps, western bias and extractive business models limit AI’s effectiveness and perpetuate historic harms, writes Mahlet Zimeta.   Chatham House

• OpenAI has quietly changed its ‘core values’ Semafor

  • unrelated, OpenAI Dropped Work on New ‘Arrakis’ AI Model in Rare Setback The Information ($)

• What’s New in the 2023 Gartner Hype Cycle for Emerging Technologies Gartner

• What is AI? (4:54)   The Prof G Show, YouTube

• I Don't Want to Be Friends with Kendall Jenner — The world doesn't need artificial companions, at least not made by billionaires.   OK Doomer

  • related, Why can't our tech billionaires learn anything new? — On Marc Andreessen’s “techno-optimist manifesto”   The Future, Now and Then, Substack

🎓 EDUCATION

• Playlab 

• Beyond ChatGPT: Princeton Language and Intelligence initiative pushes the boundaries of large AI models Princeton University

• Ruffle&Riley: Towards the Automated Induction of Conversational Tutoring Systems arXiv

• How ChatGPT is transforming the postdoc experience — Around one in three respondents to Nature’s global postdoc survey are using AI chatbots to help to refine text, generate or edit code, wrangle the literature in their field and more.   Nature

• Justin Wolfers: Assigning Homework in a World with ChatGPT (1:08:43)   Macmillan Learning

📊 DATA & TECHNOLOGY

• via patrick, How Is AI Going To Impact The SaaS Landscape Over The Coming Years with a16z GP Kristina Shen Sasstr

• A Long Way to Go: Investigating Length Correlations in RLHF arXiv

  • our takeaway: don’t bother with RLHF, just reward length

• NEFTune: Noisy Embeddings Improve Instruction Finetuning arXiv

  • NEFTune neelsjain, Github

• Why AutoGPT engineers ditched vector databases Dariusz Semba

• Introducing a New GenAI Stack: Streamlined AI/ML Integration Made Easy Docker Blog

• Clean Lab AI — Automatically find and fix errors in your real-world data. Turn unreliable data into reliable models and insights.

• The Foundation Model Transparency Index — A comprehensive assessment of the transparency of foundation model developers   Center for Research on Foundation Models, Stanford University

  • background AI Snake Oil, Substack

• What does the OpenLLM Leaderboard measure? Zeno ML Blog

• via adam, PyTorch Edge: Enabling On-Device Inference Across Mobile and Edge Devices with ExecuTorch Pytorch Blog

  • “Big deal for on device models”

• Introducing the Microsoft AI Bug Bounty Program featuring the AI-powered Bing experience Microsoft Blog

  • related, AI safety guardrails easily thwarted, security study finds — OpenAI GPT-3.5 Turbo chatbot defenses dissolve with '20 cents' of API tickling   The Register

🎉 FUN and/or PRACTICAL THINGS

• Live From the Uncanny Valley: How AI Tools Are Turning Words Into Music Pitchfork

• Texts from My Ex

  • don’t try this with current or former co-workers; the results may be disturbing

  • imessage analysis is Mac-only; whatsapp analysis is browser-based

• Greetings from Mayor Adams, generated by AI, in different languages — Eric Adams can’t speak Mandarin. But it might sound like it.   Politico

• Ikea Instructions ostris, Hugging Face

  • for a given prompt, a pictogram in the style of Ikea instructions

  • click-through for examples

• Fugees’ Pras Michél says lawyer bungled his case by using AI to write arguments The Verge

  • so genAI did what genAI does … ?

• Welcome to the AI gym staffed by virtual trainers MIT Technology Review

• Speak — Talk about anything and get instant feedback from the world’s most advanced AI language tutor.

🧿 AI-ADJACENT

• Theory-building and why employee churn is lethal to software companies Baldur Bjarnasonandand

  • great read — we could quote the whole thing