Tootfinder

LLMs are more trigger-happy than humans when it comes to launching a few nukes: #AIResearch

This paper argues that frontier models approximate tenure-level academic outputs in social science and humanities topics with "minimal engineering effort". They developed some custom agent skills to extract the qualities of individual scholars from their published works.
#LLM #AIResearch #academia

There's roughly two ways I've acquired skills in programming languages in the past: the "hard" way for writing code (e.g., "Learn Python the Hard Way"), and the "easy" way for learning to read a new programming language by skimming the language specs or leafing through a book on the topic (e.g., "The Supercollider Book").
I suppose there's a third way now for me: Reading up on software architecture design (e.g., stuff like "500 lines or less"), so that co-creation skills with large language models are improved?
For example, Yoav Rubin's article on "An Archaeology-Inspired Database" in 500 lines or less really made me think about Clojure in a new way.
Thoughts on this?
#AIResearch #Software #programming