The Nintendo is closer in time to the first transistor than to today.

Conferences are basically a way for a group of people to temporarily have a lower opportunity cost on their time.

thinking of calling this "The Illusion Illusion"

(more examples below)

Oh yeah, GANs, those were the days.

🚨 New #NeurIPS2025 paper “Training Data Attribution via Approximate Unrolling” 🚨

Introducing SOURCE: A method to understand how individual training examples influence neural net behavior, allowing us to make AI models more transparent and trustworthy!

📄 Full paper: openreview.net/pdf?id=3NaqG...

I just created a Project with a system prompt describing my interests and a doc with my publication list (titles + abstracts). Then I paste the email feed into the chat each day. Nothing fancy.

I have Claude filter my arXiv feed each day. It mostly works pretty well, except that it always hallucinates that "Studying LLM Generalization with Influence Functions" is in my feed and tells me I should read it.

Procedural Knowledge in Pretraining Drives Reasoning in Large Language Models The capabilities and limitations of Large Language Models have been sketched out in great detail in recent years, providing an intriguing yet conflicting picture. On the one hand, LLMs demonstrate a g...

Some very nice work from Cohere and UCL using influence functions to analyze math reasoning abilities in LLMs. Factual queries turn up docs containing the facts, but reasoning queries turn up similar cognitive strategies, suggesting generalization. arxiv.org/abs/2411.12580