Dominik Moritz
@domoritz.de
Visualization, data, AI/ML. Professor at CMU (@dig.cmu.edu, @hcii.cmu.edu) and researcher at Apple. Also sailboats ⛵️ and chocolate 🍫.
www.domoritz.de
www.domoritz.de
I’m getting increasingly annoyed how Ecobee cancelled their API signups to “release an improved API in the future” only to still not have done that. There is a local API through HomeKit but it’s not ideal.
October 25, 2025 at 3:43 AM
I’m getting increasingly annoyed how Ecobee cancelled their API signups to “release an improved API in the future” only to still not have done that. There is a local API through HomeKit but it’s not ideal.
Area is definitely better than radius, but not perfect. Area perception isn't itself linear. See en.wikipedia.org/wiki/Stevens.... Basically, people perceive larger circles as less large than they are.
In general, I tend to avoid size for accurate perception of magnitudes for this reason.
In general, I tend to avoid size for accurate perception of magnitudes for this reason.
Stevens's power law - Wikipedia
en.wikipedia.org
October 9, 2025 at 2:31 PM
Area is definitely better than radius, but not perfect. Area perception isn't itself linear. See en.wikipedia.org/wiki/Stevens.... Basically, people perceive larger circles as less large than they are.
In general, I tend to avoid size for accurate perception of magnitudes for this reason.
In general, I tend to avoid size for accurate perception of magnitudes for this reason.
Why don't AI people ask an AI model to help with naming? (Or maybe that's what they all did...)
October 2, 2025 at 9:53 PM
Why don't AI people ask an AI model to help with naming? (Or maybe that's what they all did...)
While I do research that is intended to live on, I don't think it would be sustainable for everyone to do the same.
It's good and necessary that people do different research with different goals. Reproducibility is a shared goal, though, that probably all research types should strive for.
It's good and necessary that people do different research with different goals. Reproducibility is a shared goal, though, that probably all research types should strive for.
October 2, 2025 at 9:15 PM
While I do research that is intended to live on, I don't think it would be sustainable for everyone to do the same.
It's good and necessary that people do different research with different goals. Reproducibility is a shared goal, though, that probably all research types should strive for.
It's good and necessary that people do different research with different goals. Reproducibility is a shared goal, though, that probably all research types should strive for.
There is a lot of great work on how viewers interpret uncertainty and how to visualize uncertainty at the last few years of the @ieeevis.bsky.social conference.
For example Hypothetical Outcome Plots (HOPs) by @jessicahullman.bsky.social. idl.cs.washington.edu/files/2015-H...
For example Hypothetical Outcome Plots (HOPs) by @jessicahullman.bsky.social. idl.cs.washington.edu/files/2015-H...
idl.cs.washington.edu
October 2, 2025 at 9:45 AM
There is a lot of great work on how viewers interpret uncertainty and how to visualize uncertainty at the last few years of the @ieeevis.bsky.social conference.
For example Hypothetical Outcome Plots (HOPs) by @jessicahullman.bsky.social. idl.cs.washington.edu/files/2015-H...
For example Hypothetical Outcome Plots (HOPs) by @jessicahullman.bsky.social. idl.cs.washington.edu/files/2015-H...
Great post. I agree that research should be abstracting away from the implementation.
I will add that protocol design itself can be a research activity and that software artifacts can be enablers of follow up research. I talked about this at the Stanford HCI seminar youtu.be/6DIIkX19ihs.
I will add that protocol design itself can be a research activity and that software artifacts can be enablers of follow up research. I talked about this at the Stanford HCI seminar youtu.be/6DIIkX19ihs.
Stanford Seminar - Build less, design more (interactive systems)
YouTube video by Stanford Online
youtu.be
October 2, 2025 at 9:36 AM
Great post. I agree that research should be abstracting away from the implementation.
I will add that protocol design itself can be a research activity and that software artifacts can be enablers of follow up research. I talked about this at the Stanford HCI seminar youtu.be/6DIIkX19ihs.
I will add that protocol design itself can be a research activity and that software artifacts can be enablers of follow up research. I talked about this at the Stanford HCI seminar youtu.be/6DIIkX19ihs.
That’s super interesting. Writing thoughts to to files could make it behave very differently from other systems making it harder to switch between models. I was kinda hoping for commoditization but I guess we get specialization.
September 29, 2025 at 11:49 PM
That’s super interesting. Writing thoughts to to files could make it behave very differently from other systems making it harder to switch between models. I was kinda hoping for commoditization but I guess we get specialization.
Reposted by Dominik Moritz
A huge thank you to co-authors @fredhohman.bsky.social, @domoritz.de, @jeffreybigham.com, @kenholstein.bsky.social, and Mary Beth Kery! This work was done during my summer internship w/ Apple AIML, and I’m thankful to work with this wonderful team :)
arxiv.org/abs/2409.18203
#UIST25 talk: Wed 11am!
arxiv.org/abs/2409.18203
#UIST25 talk: Wed 11am!
Policy Maps: Tools for Guiding the Unbounded Space of LLM Behaviors
AI policy sets boundaries on acceptable behavior for AI models, but this is challenging in the context of large language models (LLMs): how do you ensure coverage over a vast behavior space? We introd...
arxiv.org
September 29, 2025 at 3:54 PM
A huge thank you to co-authors @fredhohman.bsky.social, @domoritz.de, @jeffreybigham.com, @kenholstein.bsky.social, and Mary Beth Kery! This work was done during my summer internship w/ Apple AIML, and I’m thankful to work with this wonderful team :)
arxiv.org/abs/2409.18203
#UIST25 talk: Wed 11am!
arxiv.org/abs/2409.18203
#UIST25 talk: Wed 11am!
You're selling yourself very very short here. But that was impressively fast.
August 14, 2025 at 8:50 PM
You're selling yourself very very short here. But that was impressively fast.