weekend ai reads for 2024-03-15

Author

That AI Thing
March 15, 2024

📻 QUOTE OF THE WEEK

Dystopian fiction is when you take things that happen in real life to marginalized populations and apply them to people with privilege.

H Book Club blog (source, sorry)

 

🏗️ FOUNDATIONS & CULTURE

AI Prompt Engineering Is Dead — Long live AI prompt engineering / IEEE Spectrum

Florida Middle Schoolers Arrested for Allegedly Creating Deepfake Nudes of Classmates — In what appears to be the first criminal case of its kind, two teenage boys were charged under a 2022 Florida law for allegedly creating AI-generated images depicting middle school classmates. / Wired

How AI Could Disrupt Hollywood — New platforms and tools may allow a person to create a feature-length film from their living room. But can they really compete with the studios? / Vanity Fair

What I thought would be helpful, instead, is to survey the current state of AI-powered journalism, from the very bad to really good, and try to draw some lessons from those examples.

 

🎓 EDUCATION

The next step in higher ed's approach to AI (opinion) — The Program-Level AI Conversations We Should Be Having / Inside Higher Ed

  • … that students probably used ChatGPT to write.

 

📊 DATA & TECHNOLOGY

The first requisite for training models is acquiring compute. This seems straightforward and easy enough. However, the largest surprise turned out to be the instability of compute providers and how large variance the quality of clusters, accelerators and their connectivity were depending on the source.

  • An exceptionally helpful article to point to when considering whether to invest in a partner to train a new model from scratch: the author details all the headaches that come with managing the long-term training of an LLM outside of the confines of one of the frontier model developers (OpenAI on Azure, Anthropic on AWS (presumably), etc.). In short: hardware is hard and if you don't have folks who have experience managing training runs paired with intuitions about the settings that will lead to better performance, you're going to spend a lot of time and money and run out of runway.

AI safety is not a model property — Trying to make an AI model that can’t be misused is like trying to make a computer that can’t be used for bad things / AI Snake Oil (sorry)

via adam, The koan of an open-source LLM / Interconnects

  • A wide-ranging article on open source approaches in AI that's hard to summarize, but gives a great sense of the spirit that motivates the community.

Prompt library — Explore optimized prompts for a breadth of business and personal tasks. / Anthropic

When evaluated on the SWE-Bench benchmark, which asks an AI to resolve GitHub issues found in real-world open-source projects, Devin correctly resolves 13.86% of the issues unassisted, far exceeding the previous state-of-the-art model performance of 1.96% unassisted and 4.80% assisted. / Cognition Labs, Twitter (sorry)

 

🎉 FUN and/or PRACTICAL THINGS

Wedding Shoot Pro — Transform Your Selfies Into Wedding Photos

Big Four accountants ban grads from using AI to write job applications — Those relying on tools like ChatGPT for answers will be removed from hiring processes / The Telegraph

  • … and meanwhile, they probably still use AI to filter job applications.

Endel — Personalized soundscapes to help you focus, relax, and sleep. Backed by neuroscience.

 

🧿 AI-ADJACENT

worth it for the first footnote alone:

“Um, Keenan, are you ignoring genocide?” I hear you asking, to which my answer is, “What about you shut … up and assume that we both understand that the existence of a truly terrible thing doesn’t negate the existence of other, less-terrible things, and it’s actually alright for people to remark on the less-terrible thing without trying to ‘um, actually’ them into oblivion because they didn’t go out of their way to preempt their perspectives on the less-terrible thing by caveating in perpetuity until all communication loses meaning?” Anyway, stay tuned for my next blog post: “Hot Take: Genocide is bad.”

 

⋄