[npilk]

> You might want to use a contact form to collect structured data, so it saves you time processing requests. For example, making sure the customer provides the right identifiers for you to find them on your systems quickly, or automatically assigning queries to the right teams.

As the author alludes to, I think collecting structured data from unstructured input (e.g. inbound support emails) is a very promising real-world use case for LLMs. The goal would be to make it as easy as possible for users to send you information, and then use AI to parse what you need out of it. This would lead to less frustrated users and even increased response rates (for reasons the author mentions).

I've been playing around with this idea at https://www.semiform.ai for anyone interested.

[domdomegg]

This does seem pretty cool. I have not spent too long looking at this but a few bits of initial feedback:

1. Appreciation for an informative site: I like that your website actually explains what your product does in simple terms. So many marketing sites are impossible to parse, so it's cool that this one gets to the point quickly. Bonus points for having a live demo without a sign up.

2. Dealing with uncertainty / edge cases: One worry I'd have is that this might miss things that are relevant, or doesn't capture uncertainty well. I'd probably want a default of a 'flag for a human because this doesn't fit in the boxes well' marker by default on all forms. For example, if someone responds to the conference example with 'Sure, I'll be there on Monday and I wear a size M. Also I am in a wheelchair so will need the venue to be accessible - please let me know if it's not.', I'd want to make sure this gets flagged rather than the automated system ignoring the last part (especially as people might expect humans to read the response to an email).