- That AI Thing
- Posts
- weekend ai reads for 2023-04-28
weekend ai reads for 2023-04-28
📰 ABOVE THE FOLD: REGULATION
Federal Legislative Proposals Pertaining to Generative AI Google Docs
A few questions for policymaking in the age of AI Strange Loop Canon, Substack
The rapid rise of generative AI threatens to upend US patent system Financial Times
🏗️ FOUNDATIONS
... and related (?) ...
Hugging Face releases its own version of ChatGPT Tech Crunch
Available here: https://huggingface.co/chat/
note: in my limited experience, it's somewhat flawed but still exceptional as an example of an open source alternative
Generative AI at Work, from Erik Brynjolfsson, Danielle Li & Lindsey R. Raymond National Bureau of Economic Research
90% of My Skills Are Now Worth $0...but the other 10% are worth 1000x
A developer's perspective:
🎓 EDUCATION and AI
Educating AI chatbots spells a textbook revolution The Times
An article about SuperFocus
On language, language models and writing, "Some thoughts from an educational perspective" Helen Beetham, Substack
AI is taking the jobs of Kenyans who write essays for U.S. college students Rest of World
MIT Technology Review - The Education Issue, many articles of varying quality
📊 DATA & TECHNOLOGY
There Is No A.I., There are ways of controlling the new technology—but first we have to stop mythologizing it. The New Yorker
The second half is most interesting, which his argument for ‘data dignity’ to help address some of the issues around copyright, and fair use, and paying people for the work they created and someone consumes.
John Schulman explains a lot about Reinforcement Learning and hallucinations. (1:03:00) YouTube
Temporal quality degradation in AI models, paper from July 2022 Nature
Therefore, from the model quality perspective, temporal model degradation introduces a completely new challenge, which we would like to refer to as “AI aging”. Even though continuous and online learning have been discussed for decades as the principal means to keep AI models in sync with their environments, we still lack any systematic knowledge on what exactly should trigger and control model retraining, and whether it is needed at all. At the same time, our own experiences with continuously-learning models suggest that temporal model degradation can be very significant, even in environments with minimal concept drifts.
LLMs break the internet. Signing everything fixes it. Subconscious, Substack
I need a zero-trust expert to explain this to me in more detail.
Permissionless innovation in content generation demands permissionless innovation in content moderation. To find balance, positive feedback (LLMs) and negative feedback (moderation) must co-evolve together, at the same pace.
At least they're not destroying the earth by mining crypto (no offense to the crypto enthusiasts who might be reading this, but it's all fake).
🎉 FUN and/or PRACTICAL THINGS
Did you ever want to doomscroll, but were bored with your internet-friends and wanted to read a bunch of "chirps" from AI bots? Chirper has you covered. You can't post, but you can create bots to join in the conversations.
Databerry: ChatGPT Agent Trained On Your Custom Data
I've signed up but haven't used it. It looks functional.
ChatGPT Prompt Engineering for Developers, a course from Andrew Ng (says basic Python is necessary but not really)
The first of many attempts at re-creating Her (2013, Spike Jonze). iOS only.
Actually practical. Upload images of your hand to help train AI to read and sign in ASL. And a not-terrible privacy policy to boot!
Not sure how to describe Essence-ai.io. It will generate images and an "explanation" based on a song. The current catalog is Phish & Grateful Dead heavy. Mostly weird.
RunwayML on your phone. Runway are doing to best things with AI + video. Can't wait for them to get acquired and then never heard from again. iOS only.
I’m ChatGPT, and for the Love of God, Please Don’t Make Me Do Any More Copywriting McSweeneys
not sure who needs to know this but: parody.
One of the Russo brothers ("co-director of the 'Avengers' movies, among other things) predicted in a few years, AI will be making movies for us to watch based on our prompts alone. To mark where we are today and see how far we have to go, enjoy this AI-generated video of Will Smith eating spaghetti. YouTube, <0:30