suhr

Sharing this really cool paper of ours that will appear at EMNLP in November: arxiv.org/abs/2606.319...

DigitalCoach: Communication and Grounding Gaps in Human and Agentic Computer Use Coaching arxiv.org

(wait are you at NYU? maybe I will email you when it comes time to design my next seminar and we can do something together... I do a decent amount of stuff with games, like arxiv.org/abs/2512.03381)

Characterizing Language Use in a Collaborative Situated Game arxiv.org

Image generation models typically don't use the same algorithms/implementation as LLMs, but generative models of music do (e.g., arxiv.org/abs/1809.04281), so I like to use these models as examples of this point!

Music Transformer arxiv.org

@ali-alkhatib.com puts it nicely ali-alkhatib.com/blog/definin...

Ali Alkhatib: Defining AI ali-alkhatib.com

Is anyone studying the "cognitive load" of reading LLM-generated text? The extra interpretative labor that presenting such text asks of us?

Huh, I just realized that one meaningful difference between some kinds of LLM use and other kinds of LLM use might be whether you're afraid of its outputs being watermarked.

I'm somehow reviewing the "neuroscience and cognitive science" track at NeurIPS and it's incredible what bullshit AI/ML researchers can come up with when they read Wikipedia articles about the (still mostly-shaky) foundations provided by psychologists

See, the real creation of a "rogue AI" already happened. It happened as soon as people anticipated its existence, which was long ago. First in allegories, then in people who crave to be a subject of of some mystical force they refuse to understand. These technologies aren't wild animals.

how much of "AI xrisk" panic is actually just a manifestation of anxiety about how fragile our social and material infrastructure might be, now that so much of it is Computer?