These 7 things happened in AI today

7 top AI stories today. I can’t decide if the first story is funny, horrifying, or both.

Cheers,

Marshall

Martin Shkreli Launches AI for Medical Advice
The controversial figure Martin Shkreli (Wikipedia) has launched a medical chatbot called DrGupta.ai. Critics online, including Dr. Sasha Luccioni of HuggingFace, were outspoken in opposition of the use of LLMs in medical advice due to accountability issues. Luccioni also rejects the claim that AI like this can aid the economically disadvantaged, calling it a myth. The chatbot is available online and can be asked any question, from the extremely juvenile to the serious (I assume). [via @SashaMTL]

Scientific discovery in the age of artificial intelligence
A group of 30 authors published a paper in Nature today about the uses of AI to speed up scientific discovery through (1) theory exploration, (2) experiment design, and (3) data analysis. May that be an inspiration to us all. [Scientific discovery in the age of artificial intelligence]

Simon Willison Launches LLM Plugin for Llama 2 Operation on a Mac
Developer Simon Willison introduced a plugin for his LLM utility, adding compatibility with Meta AI's commercially licensed Llama 2. [Run Llama 2 on your own Mac using LLM and Homebrew]

Patterns for Building LLM-based Systems & Products
Amazon scientist Eugene Yan has written a very long and widely-praised blog post on LLM integration patterns. He says LLMs can be integrated into systems using seven patterns: (1) evaluations, (2) retrieval-augmented generation, (3) fine-tuning, (4) caching, (5) guardrails, (6) defensive UX, and (7) user feedback. He also writes about an emerging trend of automated evaluations through a robust LLM, a cost-effective alternative to human evaluations despite potential biases. [Patterns for Building LLM-based Systems & Products]

Jupyter Notebooks Add Generative AI
The popular Project Jupyter, has added generative AI to its notebooks, enabling users to generate and explain code, fix errors, and query local files. It links Jupyter with large language models from providers like AI21, Anthropic, AWS, Cohere, and OpenAI, using LangChain for model support. Jupyter AI emphasizes responsible AI and data privacy, only interacting with a language model upon request and never sharing data without permission. [Generative AI in Jupyter]

Meta AI Launches AudioCraft for Generative Audio
Meta AI launched another open source code base, this one designed to cater to all generative audio needs, including music and sound effects. The platform also includes AudioGen for text-to-sound generation and MusicGen for text-to-music generation. The project page does not appear to have a hosted demo. [AudioCraft: AI research for audio]

New Course on Evaluating and Debugging Generative AI
DeepLearning.ai has launched a free one-hour program in partnership with Weights and Biases. Participants will learn to use Machine Learning Operations tools and the Weights & Biases platform. [Evaluating and Debugging Generative AI]

That’s it! Feel free to share this email full of summaries with a co-worker or loved one. I’d love it if you did.

thanks, Marshall