iiinova — where your digital self keeps evolving

move37.bsky.social @move37.bsky.social · 24/08/2026

User involvement? I don't think it's contractually specified anywhere to involve users in things like requirements elicitation or pre-deployment testing etc.

001

move37.bsky.social @move37.bsky.social · 15/06/2026

Are there valid ways to test this unreliability? Either by recognizing pre-transcribing that this speech is in a category that will be unreliable or by quantifying the uncertainty of the transcript itself? (not a computational linguïst, sorry if this is an obvious question)

010

move37.bsky.social @move37.bsky.social · 13/02/2026

It's not ideal (esp. when an AI interviewer is as verbose as you note), but it's an improvement compared to not interviewing/observing them at all. This type of situation may occur more frequently in e.g. organization studies, where universality is less of a concern than cognitive science (/end)

010

move37.bsky.social @move37.bsky.social · 13/02/2026

For example, when studying an organizational change, there's often a limited number of stakeholders you can interview/ observe, but at the same time you usually have a reasonable (although perhaps incomplete) idea of what stakeholders might be underrepresented, for which AI might be a surrogate (2/)

100

move37.bsky.social @move37.bsky.social · 13/02/2026

I read the TiCS paper and pretty much agree with all of your critiques. There's one nuance I'd like to offer though and that is that sometimes it is less about the universality of the sample, but more about the representativeness of the sample for a much more bounded population. (1/)

100

move37.bsky.social @move37.bsky.social · 31/01/2026

The mansplaining you refer to seemed to me more 'explaining for his audience' (which likely is less expert on AI)? I've been in multiple prep meetings with journalists/panel mods where they mentioned they would do that before coming to me with their Q. Not sure if you had such a meeting though.

030

move37.bsky.social @move37.bsky.social · 10/05/2025

Indiscriminate application of AI in govt services deserves critique, but being able to provide information in 70 languages instead of only English and Spanish seems like a legitimate application, especially since this was the original use-case that transformers were developed for and tested against?

100

move37.bsky.social @move37.bsky.social · 29/04/2025

"We're experimenting in the backs of pupils" makes it sound scary, but it's nothing new. How do you think educational innovations pre-AI got into schools?

000

move37.bsky.social @move37.bsky.social · 29/04/2025

He regularly post them here, so you can find them on his timeline. His timeline on Twitter goes much further back with additional studies.

040

move37.bsky.social @move37.bsky.social · 28/04/2025

The "GM having P&L (Pillage &Loot) responsibilities"was a nice touch too 😁

000

move37.bsky.social @move37.bsky.social · 21/02/2025

The Discoverers by Daniel Boorstin. Wonderful appreciation of human ingenuity and drive.

010

move37.bsky.social @move37.bsky.social · 13/02/2025

I wouldn't trust this last one to be honest. Very long journal name with lots of buzzwords, journal website looks very dodgy: barely any info about the journal itself and the 'archives' for the 10 volumes only go back to December 2024. Probably a pay-to-publish honeypot.

060

move37.bsky.social @move37.bsky.social · 10/02/2025

At the risk of running ahead of the actual launch, but will SHADES have a user-friendly interface? I have many people in my organization that are actively working with LLMs who are not coders, but are concerned about LLM bias. SHADES could be a great tool for those people!

010

move37.bsky.social @move37.bsky.social · 25/01/2025

On one hand I very much agree that that is the better question to ask. On the other hand there are limits to how far this will get you: sufficient transparency ex ante isn't always possible (e.g. the telephone) and trustworthiness also comes from the experience of using it and succeeding/failing?

010

move37.bsky.social @move37.bsky.social · 21/01/2025

But would that then mean that in your opinion all textual/verbal experiments on theory of mind are invalid and only behavioral experiments count?

210

move37.bsky.social @move37.bsky.social · 08/01/2025

Thanks! A lot of these points resonate highly with the work I do implementing Responsible AI at my university. One aspect I think deserves a bit more emphasis is how to do this in more or less real-time during development, in addition to auditing systems in production. Any thoughts?

000

move37.bsky.social @move37.bsky.social · 19/12/2024

Thank you! I really appreciate this motivation and line of reasoning. Swiss Army knives are nice and useful, but a proper tool gets the job done better and faster

010

move37.bsky.social @move37.bsky.social · 05/12/2024

I wish I knew. My Discover feed is at least 80% posts from people that I follow, not much to discover there...

010