Shadab's Bluesky: @namer.bsky.social

We look at a common task in Augmentative and Alternative Communication (#AAC) where keywords are expanded into full sentences and see how using different emotion representations affects how LLMs generate sentences.

In this work, we introduce a new human evaluation of #LLMs called Representation Alignment. Representation Alignment looks at the similarity between how LLMs use conceptual representations and how humans interpret them.

You can check out the pre-generated narratives based on Marvel movies here: what-if-game.github.io

Runsheng (Anson) Huang's (who worked with me at Penn) paper "WHAT-IF: Exploring Branching Narratives by Meta-Prompting Large Language Models" is a system that turns existing plot summaries into engaging branching interactive fiction.

Preprint: arxiv.org/abs/2412.10582

Patty Delafuente & Arya Honraopatil's paper "Does Reasoning Help LLM Agents Play Dungeons and Dragons? A Prompt Engineering Experiment" looks at using reasoning models like DeepSeek for generating actions of D&D players based on game state. Turns out this isn't something models can reason about!

Or maybe your dreams as a janitor will lead you to be the captain of your very own garbage ship!

Northwestern just announced that the university will meet the funding needs of any research that is being impacted by stop orders. LET'S GOOOOOOOOOOOOO! #edusky #academicsky

wgntv.com/evanston/nor...