At this point I feel like we understand pretty well what's going on with LLMs:

- Outputs are roughly equivalent to kernel smoothing over positional embeddings (arxiv.org/pdf/1908.11775…)
- The learned computation model is *probably* bounded by RASP-L (arxiv.org/pdf/2310.16028…)
-

Paul Yamazaki of City Lights reflects on the joys of running an independent bookstore: “At a great store you can look at twelve well-selected, serendipitous linear inches and find a universe.”
lithub.com/paul-yamazaki-…

This, more than any order unit of comparison between natural and artificial intelligence, makes one think. GPT-4 consumed 62 million kWh. The typical human brain over the *total* period of a 100 year lifetime is estimated to consume 1 million kWh. Surely, we aren’t there yet?! 🤔

It is saddening to observe the similarities between what is described in this open letter from UC Berkeley faculty and the situation in many major UK universities.
ucbfa.org/2024/04/an-ope…

The Breakthrough of the Year @AAAS Bhaumik Prize goes to the 2 scientists who pushed/fought to get GLP-1 drugs to treat obesity (which took decades!)
aaas.org/news/innovator… by Meagan Phelan, with the whole, fascinating story
pleased to be a member of the selection committee

“think ad targeting but for killing”!

Michael Genesereth on giving teens tools for thought news.stanford.edu/report/2024/04…

Philipp Staab’s exploration of how tech giants operate like the colonising East India Company offers a nuanced critique of the fast-developing digital economy.

Tap here for our review: on.ft.com/3TVzbiB

Sail far. Sail wide. Sail on. ⛵

#inspiration #HCLDR #leadership #pinksocks

When large AI models become proficient at narrow tasks, there's a tendency to oversimplify and reduce a discipline to a one-liner description that encapsulates the model's output. It's as if we're trying to solve for the first sentence in a Wikipedia entry: filmmaking is reduced

“Good writing is about having something interesting and original to say. Generative AI tools might provide technical help, but they are no substitute for your unique perspective.' nature.com/articles/s4225…

Elon Musk and Sam Altman may not agree on much of late, but do agree AI is built on strong math foundations, including algebra and calculus, applauding University of California for recent clarifications on math requirements for admission.

Many industry leaders signed:

mathmatters.ai

“…developers using GitHub’s Copilot AI assistant are 55% more productive. The catch? Copilot yields code with security bugs and design flaws 40% of the time.” 🤔 fastcompany.com/91045570/thank…

📢 Call for Papers! ⚡️🚗

Emerging Trends and Concepts for Future #Automotive Electronics and Software

Following a The Alan Turing Institute workshop, we're delighted to collaborate on this special collection.

🗓️Deadline: 31 July 2024
👉How to submit: bit.ly/3TkBY4U

S Ramamoorthy

Important observation. Easily and often forgotten, unfortunately.

Poetry Will Not Optimize; or, What Is Literature to AI? | American Literature read.dukeupress.edu/american-liter…

I’m pleased to be involved in these, and look forward to working with Themis Prodromakis within the AI for Productive Research and Innovation in Electronics Hub (APRIL), and with Sotirios (Sotos) Tsaftaris in The EPSRC AI Hub for Causality in Healthcare AI with Real Data (CHAI).

To lawyers using AI: Caveat Emptor

'Hallucinating Law: Legal Mistakes with Large Language Models are Pervasive'

hai.stanford.edu/news/hallucina…

This is a good visualization of the (obscene) size of the data and compute for LLMs. This feels like a very inefficient information extraction machine. Scale is not enough…