Tooling

Wired's Steven Levy details how eight Google employees invented the "transformer" concept, leading to a recent tech breakthrough. via kottke.org

Ken Kantzer shares practical lessons learned from using OpenAI's models, including a simple token estimation method and challenges with structured data extraction.

via Simon Willison's Weblog

Molly White's newsletter post explores the potential benefits and significant drawbacks of AI tools, questioning if AI's benefits outweigh its ethical and societal costs. via @citneedednews

Mashable calls Adobe's new AI video editing tools "absolutely terrifying" for their potential to make it harder to tell what's real in videos. via Mashable.

Researchers at UPenn have developed a system that uses AI to generate 3D environments from text prompts, similar to the technology seen in Star Trek.

Researchers put ChatGPT to the test, measuring its ability to mimic user behavior in tree testing on the IRS website and found near-perfect task completion, surpassing human participants. via MeasuringU.

A new 2.7B parameter language model named BioMedLM, now available on Hugging Face, shows strong performance on biomedical question answering after fine-tuning. via Hugging Face.

A new study found that law students using GPT-4 didn't improve work quality but did work faster, with the AI being most helpful to lower-performing students. via Marginal REVOLUTION.

Matt Webb explores the future of personal AI agents and how new search engines might help them locate and utilize tools like apps. via Interconnected

TnT-LLM, a new framework, uses large language models (LLMs) to automate text mining tasks like generating label taxonomies and building text classifiers. via Hugging Face.

Zach Seward discusses how AI can be a useful journalism tool for handling large amounts of data, but also how it's being used to replace human writers. via kottke.org

MIT, Harvard, and MIT researchers introduce a new method for conducting economics research and modeling human behavior using a system devoid of human input. via Marginal REVOLUTION