Post

Daniel Abraham
@abraham.bsky.social

So Muse is a software that asks for all your personal data, and then does *stuff* with an unknown, non-zero error rate? If my assistant accidentally pulled forms for a gym that I 1) didn’t belong to and 2) didn’t live in the same state, I’d assume they were ill. www.wsj.com/podcasts/the...

wsj.comOur Reporter Tried Meta’s Viral AI Agent. Should You? - The Journal. - WSJ PodcastsCome see The Journal Live! Tickets for our show in Dallas are on sale now. Muse is Meta’s new AI agent, and it’s a hit. With more and more people using this type of technology to help manage their tim...

1 repost25 likes

Django Wexler
@djangowexler.bsky.social

This seems like such a bad area to use AI because the potential gains are very small ("I didn't have to fill out that form!") and the costs of failure are very large ("It transferred all my money to hackers.") so to make the cost-benefit work out, the error rate has to be *very very low*.

Sean Eric Fagan
@kithrup.bsky.social

LLMs are pretty good for intent processing. But it's how you use that is key. (eg, you can have a table that is something like: if phrase means "show me a movie" then {} else if phrase means "what time is it" {} etc., and the LLM can allow for a *wide* variety of actual phrases to be used.)