Ahana (she/her)

✨ The last day kicked off with an amazing talk by @katjahofmann.bsky.social
"World and Human Action Models for Gameplay Ideation" 🎮🤖

Exciting vision from the Game Intelligence team @msftresearch.bsky.social

I am in Vancouver at ICML, and tomorrow I will present our newest paper "Partially Observable Reinforcement Learning with Memory Traces". We argue that eligibility traces are more effective than sliding windows as a memory mechanism for RL in POMDPs. 🧵

After all these reports of authors adding language instructions for LLM reviews in their papers I wanted to check this myself and I downloaded the .tex source from one of these papers.

Here is an example.
(I will not share the identity of the paper)

I SENT A WHOLE FUCKEN SPOON THRU THAT THING AND NOTHING HAPEPNED AT ALL! SO WHAT DA FUCK!!! WHAT ELSE HAVE WE BEEN DOING OUR WHOLE LIFE'S THAT WAS A TOTAL LIE AND WHOSE BEEN KEEPING THE SPOONS OUT OF THE MICRO WAVE AND HOW ARE THEY PROFITTING FROM IT !!!!!!! AND DA TEXT SAYS "ACCIDENTALLY PUT A DAMB SPOON THRU THE MICRO WAVE AND NOTHING BAD HAPPENED" AND A SKELITEN IS HOLDEN THERE SPOON AND ITS FINE I GUESS . I DONT KNOW - DASHARE.ZONE ADMIN

WHAT ELSE DID THEY LIE ABOUT - dashare.zone ADMIN

Join us for Nneka's presentation tomorrow! Last talk before the summer break.

new preprint with the amazing @lviano.bsky.social and @neu-rips.bsky.social on offline imitation learning! learned a lot :)

when the expert is hard to represent but the environment is simple, estimating a Q-value rather than the expert directly may be beneficial. lots of open questions left though!

new work on computing distances between stochastic processes ***based on sample paths only***! we can now:
- learn distances between Markov chains
- extract "encoder-decoder" pairs for representation learning
- with sample- and computational-complexity guarantees
read on for some quick details..
1/n

Government officials are letting AI do their jobs. Badly Offloading government responsibilities to AI can encourage discrimination, give wrong advice, and limit access to valid claims of asylum.

"The chatbot responded that it was perfectly okay for landlords to discriminate based on whether those potential tenants need rental assistance"

From "Government officials are letting AI do their jobs. Badly," by @emilymbender.bsky.social & @alexhanna.bsky.social

thebulletin.org/2025/05/gove...

Disabled people have been pleading with folks to show solidarity for years.

We saw the rising tide of eugenics the moment everyone declared Covid over & decided to leave us by the wayside.

Pandemics give rise to fascism. What we’re seeing is the result of allowing disabled folks to be left behind

found a cool game theory book
arxiv.org/abs/1512.06808

oofe I'm instantly hooked. Great find!

Trying to get my head round how this could make it to print. Many warning signs for all of us

Best of luck!!

In the year of our Lord 2025... we are still putting on makeup on women without their consent.
It's truly like feminism never happened 🙅‍♀️
(I have been ranting about this ever since people started using it to demonstrate GANs back in the 2010s, I can't believe it's still an "acceptable" task in AI!)

thank you! I really appreciate you taking the time <3

Why I don't write in my mother tongue an attempt to untangle my feelings about my mother tongue Bangla

I have started a substack lately on my random thoughts, if anyone's interested <3

Books on the bed

Singapore book haul! (+ St Jordi)

I do not necessarily agree with your first post, but this is really really disgusting and completely unwarranted. Hope you’re instantly blocking people like this, what the hell. Sending a lot of love and support <3

our work ~at~, sorry it's 2 am

Sadegh Talebi explaining the poster.

Poster png

Had a lot of fun presenting our work at "Offline RL in Regular Decision Processes: Sample Efficiency via Language Metrics" at #ICLR2025, with my co-authors Alessandro Ronca and Sadegh Talebi!

check out our paper here: openreview.net/forum?id=EW6...

"It's wonderful watching the well of knowledge being poisoned in real time".
📷 Emily Gorcenski

First draft online version of The RLHF Book is DONE. Recently I've been creating the advanced discussion chapters on everything from Constitutional AI to evaluation and character training, but I also sneak in consistent improvements to the RL specific chapter.

rlhfbook.com

A mid-week self reminder

A relatedly bad idea is working on research you don't deeply believe in for pragmatic reasons. You start with "oh, it'll be a quick low-hanging fruit project" and 9 months later you're still working on it and in despair

Took a weekend off to travel to Bruges. It was beautiful :O

Mark your calendars, EWRL is coming to Tübingen! 📅
When? September 17-19, 2025.
More news to come soon, stay tuned!

Thought-canceling headphones

Beautiful.

Turing Award Goes to A.I. Pioneers Andrew Barto and Richard Sutton Andrew Barto and Richard Sutton developed reinforcement learning, a technique vital to chatbots like ChatGPT.

Congrats to this year's Turing award winners! www.nytimes.com/2025/03/05/t...

Incidentally, if you'd like to hear from them, we know a place they've given / are giving keynotes

Made a small drawing in memory of my partner’s cat :’)