My paper "Role Play with Large Language Models", co-authored with Kyle McDonell and Laria Reynolds, is out in Nature, and is free-to-read:
https://nature.com/articles/s41586-023-06647-8

'The Bayesian Learning Rule', by Mohammad Emtiyaz Khan, Håvard Rue.

http://jmlr.org/papers/v24/22-0291.html

#bayesian #gradients #gradient

Join us at #ICCV workshop on uncertainty quantification

Elon Musk, owner of X, the platform formerly known as Twitter, tells people on X, the platform formerly known as Twitter, to stop calling X, the platform formerly known as Twitter, “X, the platform formerly known as Twitter”

I have been avoiding Zoom for teaching, professional and personal meetings as much as I could since the start of the pandemic, using and recommending Jitsi instead.

The new updated terms of Zoom just give you new reasons if you were accepting Zoom so far.

Very clever — the German ad says "protect wildlife before it's too late."

HT WWF Germany

Talk by Amy zhang on Dual RL: Unification and New Methods for Reinforcement and Imitation Learning

Going to work at the University of Sydney is incredible. Here, a Kookaburra, fiercely and proudly looking at its nearby nest (a hollow in a tree of Victoria Park, two minutes from my building). #USyd #OwnPic #BirdsofOz

Excited to be at #ICML2023 in Hawaii 🌺 Excited to give a keynote on Tuesday entitled ‘Machine Learning with Social Purpose’ 🎈We’ll be covering everyone from weather forecasting, statistical testing, sociotechnical systems, and global community building.

I am attending ICML at Hawaii & will be around to chat
- July 23 attending AABI symposium whole day
- July 25, 26, 27 poster sessions 10-3pm
- July 29 at our workshop on duality principle https://dp4ml.github.io/

If you want to meet, let me know!

For those who care, GPT-4 details have leaked (apparently an open secret in Silicon Valley circles)

https://twitter.com/soumithchintala/status/1671267150101721090?s=20

new preprint

"ReLU to the Rescue: Improve your On-policy Actor-Critic with Positive Advantages"

shockingly simple changes to A3C can give a cautious RL algorithm more effective than PPO

in some settings, just adding a ReLU is enough!

💻github.com/anndvision/vsop
📄arxiv.org/abs/2306.01460

with @Yarin, Gunshi Gupta, Angelos Filos, Chris Lu, and Jakob Foerster

I hope they solve the confabulation/hallucination issue. I used Bing Chat (aka GPT-4) to get recommendations for the NCAA tournament and it really liked to make up facts. I would have lost a lot of money if I gambled on GPT-4 sports recommendations.

A lawyer is in big trouble after using ChatGPT to do legal research

https://storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.nysd.575368.32.1.pdf

Day 2 of #iclr2023 kicks off with Masashi Sugiyama on distribution shift adaptation

Here, I made you a Slack emoji for when you're talking about GenAI

ChatGPT ties itself in knots to avoid having professors be female.

Source: https://twitter.com/ndyjroo/status/1649821809154613248?s=61&t=Ugdi4XBKf_2ovJ1y9hKs4w

This is real-world XKCD

See full thread: https://twitter.com/rachel_l_woods/status/1649790435533922307?s=61&t=Ugdi4XBKf_2ovJ1y9hKs4w

Gary Marcus: "AI is hitting a wall"

AGI boosters: "No No AI is not hitting a wall"

Sam Altman:

RT @CohereAI
What could you build if you had the embeddings of ALL of wikipedia?

The Embedding Archives: Millions of Wikipedia Article Embeddings in Many Languages
https://hubs.li/Q01Mg6_C0

We’re publishing ~100 million embedding vectors, covering Wikipedia in 10 languages. Get them now!