- That AI Thing
- Posts
- weekend ai reads for 2024-01-05
weekend ai reads for 2024-01-05
📰 ABOVE THE FOLD: NEW YORK TIMES LAWSUIT
The Times Sues OpenAI and Microsoft Over A.I. Use of Copyrighted Work — Millions of articles from The New York Times were used to train chatbots that now compete with it, the lawsuit said. New York Times
the lawsuit [PDF]
Open AI may have changed things in light of this lawsuit but our own testing couldn’t replicate this even with prompting that was beyond “minimal”
that basically covers all LLMs today
also, isn’t “LLM models” is like “PIN numbers”? — stop it
point: If you want to understand why the Times case has a near zero probability of winning, then read this thread. Daniel Jeffries, Twitter (sorry)and
counterpoint: Things are about to get a lot worse for Generative AI Marcus on AI, Substack
the image examples are more damning but not part of the Times lawsuit
OpenAI’s news publisher deals reportedly top out at $5 million a year The Verge
What Do AI Companies Want With the Media? Intelligencer, New York Magazine
📻 QUOTE OF THE WEEK
It takes someone who really knows the guts of what you are doing to appreciate a black triangle.
“Rampant Coyote” (source)
🏗️ FOUNDATIONS & CULTURE
The road ahead reaches a turning point in 2024 — This year signaled the start of a new era. Here’s why I believe next year is an opportunity to shape the world’s next chapter for the better. Gates Notes
What kind of bubble is AI? Pluralistic (Cory Doctorow)
solid closer
AI and Mass Spying Schneier on Security
related, Chatbot users should not share private information with software, expert warns — Prof Michael Wooldridge said complaining about personal relationships or expressing political views to artificial intelligence was ‘extremely unwise’. Independent
maybe obvious? but maybe not since there’s a whole article about it
An Idiot's Guide to Hugging Face Very Explainable, Substack
Explaining ChatGPT to Anyone in <20 Minutes Deep Learning Focus, Substack
all these topics have been covered elsewhere, but another holistic view of all the piece-parts
AI image training dataset found to include child sexual abuse imagery — Stanford researchers discovered LAION-5B, used by Stable Diffusion, included thousands of links to CSAM. The Verge
Big Tech outspends venture capital firms in AI investment frenzy — Microsoft, Google and Amazon crowd out traditional Silicon Valley investors in blockbuster deals with AI start-ups in 2023 Financial Times
“The Universe Wants Us to Take Her Clothes Off” With Grimes Palladium Magazine
warning: nsfw image at link; image-free pdfand
Have you heard the word Internal Exile, Substack
AI is going great for the blind. Robert Kingettandand
do read the whole thing
🎓 EDUCATION
via deven, Cheating Fears Over Chatbots Were Overblown, New Research Suggests — A.I. tools like ChatGPT did not boost the frequency of cheating in high schools, Stanford researchers say. New York Times
How GenAI can turn an autobiography into an interactive Black history lesson Tech Crunch
A Cross-Disciplinary Examination of the Instructional Uses of ChatGPT in Higher Education Research Gate
Technology Students in Africa, Bolstered by ‘Grassroots AI’ — Amid South Africa’s perfect storm of few universities, racial inequities and economic disparities, AI students find support in a not-so-grassroots grassroots movement. Inside Higher Ed
Zelma — Education data a question away
After Transforming a College With Online Offerings, a President Steps Down to Tackle AI Ed Surge
Student Perspectives on Using AI (7:09) Educause Exchange
transcript at the link
My Predictions for AI-Powered Higher Education Kiran Kodithala, Medium
plus three more at the link
📊 DATA & TECHNOLOGY
Understand and Exploit GenAI With Gartner’s New Impact Radar Gartner
“User-in-the-Loop AI” is 6-8 years away
How To Choose Your LLM AWS Community
Mixtral of experts — A high quality Sparse Mixture-of-Experts. Mistral News
related, Mixture of Experts Explained Hugging Face Blog
Will AI finally take real-time data mainstream? — Streaming pipelines have always been costly and difficult to implement. But rapid experimentation in AI means the wallets are suddenly open to revisit it. Supervised ($)
AI agents help explain other AI systems — MIT researchers introduce a method that uses artificial intelligence to automate the explanation of complex neural networks. MIT News
How to Build the Right Team for Generative AI Human Loop
emphasis ours
Building Trustworthy Generative AI Deloitte AI Institute
the report [PDF]
Improving Text Embeddings with Large Language Models arXiv
new frontier for RAG?
the researchers asked the LLM three things:and
really neat
Apple’s AI research signals ambition to catch up with Big Tech rivals — Paper suggests a focus on running the latest technology directly on an iPhone rather than via the internet Financial Times ($)
the paper: LLM in a flash: Efficient Large Language Model Inference with Limited Memory arXiv
loosely related, MLX Community — A community org for model weights compatible with mlx-examples powered by MLX. Hugging Face
Interviewing Tri Dao and Michael Poli of Together AI on the future of LLM architectures (35:46) Interconnects
Helix — Build and optimize text and image AI for your needs. Train, fine-tune, and generate from your data. Deploy in your cloud — or use ours.
surprisingly good; medium-to-high technical skills necessary
Phi-2: The surprising power of small language models Microsoft Blog
🎉 FUN and/or PRACTICAL THINGS
Eternal Jukebox — generate a never-ending and ever changing version of the song
“Pick a song”, pick a song, then click “Play”
great for background music when you want to have the same song on repeat but not quite; decent visuals too
Emoji Translator — Translate any text into emojis, and vice versa Go Space Dev, Hugging Face
What I Wish Someone Had Told Me SamAltman.com
Microsoft’s Stuffing Talking Generative AI Into Your Car — “Generative AI is going to add an emotional layer to interaction with the car.” The Byte
no
ai marketing-speak is out of control
Trends for 2024 Google Drive
200 or so predictions for 2024 from everyone from Accenture to YouTube
Suno — Make a song with Suno
this is the worst that generated music is going to be for the rest of your life
AI can now tell if your Louis Vuitton handbag is fake Business Insider
finally!
AI can tell which chateau Bordeaux wines come from with 100% accuracy — A machine-learning algorithm was able to tell which estate 80 Bordeaux red wines came from with 100 per cent accuracy by assessing their chemical signatures New Scientist
finally!
a new search engine, like Perplexity
infographic with gems about each minute, such as “the average person produces 102MB of data” and “instagram users send 694k reels via dms”
Tracking AI — Monitoring Bias in Artificial Intelligence Chatbots
Mistal is closest to “centrist”
Rite Aid Banned from Using AI Facial Recognition After FTC Says Retailer Deployed Technology without Reasonable Safeguards — FTC says Rite Aid technology falsely tagged consumers, particularly women and people of color, as shoplifters; Ban will last five years Federal Trade Commission
News Déjà Vu News Déjà Vu, Github
Audiobox — Where anyone can make a sound with an idea Meta
impressive generative audio, including replicating your voice, and describing a voice that then speaks words you tell it to
Bricking it: Do you actually own anything digital? — From ebooks, to videos and software, the answer is increasingly no The Register
Mickey 1928 Pclanglais, Hugging Face
not quite there; generates images of Mickey with white gloves, which is a version of Mickey not yet in the public domain; the creator acknowledges as much which is nice:
and for fun: Steamboat Willie (7:46) Internet Archive
🧿 AI-ADJACENT
1000 Matches (1:17:24) Swimming, YouTube
any 2-minute segment will charm you