weekend ai reads for 2023-07-21

July 21, 2023

I warned you guys in 1984, and you didn't listen.

James Cameron, writer & director of Piranha II: The Spawning (source)

Meta and Microsoft Introduce the Next Generation of Llama Meta
- Open source LLM that can run locally to compete with ChatGPT, Claude, and a thousand others.
- Just briefly using it, two things jumped out: it’s either worse or has stricter guardrails (e.g., “tell a joke about…” prompts), and it’s fast.
- try it here: Llama Chat
- via adam, Llama technical paper
- related: ChatHub — Use ChatGPT, Bing, Bard, Claude and more chatbots simultaneously
The hottest new job is “head of AI” and nobody knows what they do Vox
- related: 59% of orgs lack resources to meet generative AI expectations: Study Venture Beat
The $1 billion gamble to ensure AI doesn’t destroy humanity — The founders of Anthropic quit OpenAI to make a safe AI company. It’s easier said than done. Vox
- related: AI Is an Existential Threat—Just Not the Way You Think — Some fear that artificial intelligence will threaten humanity’s survival. But the existential risk is more philosophical than apocalyptic Scientific American
In the Age of A.I., Tech’s Little Guys Need Big Friends New York Times
Actors vs. AI: Strike brings focus to emerging use of advanced tech NBC News
- related: Actors say Hollywood studios want their AI replicas — for free, forever The Verge
What Would You Pay for Personalized AI? — The math for the $1T+ Personal AI market The Deload
Are Large Language Models a Threat to Digital Public Goods? Evidence from Activity on Stack Overflow arXiv

Data and Compute Are the Ultimate Flywheel — What’s valuable—and why—in an AI-driven world Every Media
- Overview of the changing data loops, and impacts on, well, the world
PoisonGPT: How we hid a lobotomized LLM on Hugging Face to spread fake news Mithril Security
- Oh, great.
Announcing LangSmith, a unified platform for debugging, testing, evaluating, and monitoring your LLM applications LangChain
- Currently in closed beta; looks like it will make shipping AI-informed products easier, for better or worse (see above).
AI Agents that “Self-Reflect” Perform Better in Changing Environments Stanford University
Lost in the Middle: How Language Models Use Long Contexts arXiv
- It turns out that ChatGPT performs less well in terms of retrieving information if it’s contained in the middle of a document. Other LLMs like Claude perform slightly better.
Ecosystem Graphs Stanford University
- Long inventory of models, datasets, and apps
- related: via adam, Constellation: An Atlas of 15,000 Large Language Models

Viola the Bird Google Labs
- Charming “ML” experiment where you control a bird playing a cello
Artists are losing the battle against AI. Glaze, a tool that's found a way to trick algorithms, is giving them a fighting chance. Business Insider
Fable unveils Showrunner AI to create South Park-like TV shows with you as the star Games Beat
Introducing NotebookLM — An AI-first notebook, grounded in your own documents, designed to help you gain insights faster. Google Blog
- Currently has a waitlist; after just a little testing, perhaps we can retire our Zettelkasten note-taking approach.
Stable Doodle — Transform your doodles into real images in seconds Stability AI
Scientists just used A.I. to map a fruit fly’s brain. Here’s why it’s a ‘turning point in neuroscience’. Fortune
Custom instructions for ChatGPT. OpenAI
- Currently available for Plus customers.

‘Sharenting’ (2:50) Deutsche Telekom, YouTube
- Intense depiction of the risks of sharing your kids’ information online
Amazing Commercial Featuring the French National Football Team kottke.org
- Even if you don’t have any interest in either football/soccer or France, this is great