I know that sounds bad but it’s true, at least I my experience.
I think the issue here is, how do you keep this model up to date ? Maybe that’s a trivial thing to do ?
Associate a timestamp with documents and use that to decay the importance. Assign higher importance to key policy and onboarding docs, company-wide emails, etc.
The PM/EM for a team will become responsible for curating a team's additional training data.
Let the system itself regularly suggest information to prune. Have an internal team analyze this and reach out to the relevant stakeholder(s) to confirm when unclear.
If Slack, Microsoft, and Google aren't re-shuffling their priorities to build this right now, they're crazy.
I’m honestly not to excited about where tech is going lately.
I guess we’ll soon be expected to where mind reading devices to keep the models up to date ?
I know that I've wasted hours searching Gmail, docs, Slack, and Jira, or just given up entirely before finding what I want.
Think of an LLM for biz docs as just being incredibly good search. Where you don't have to know the search terms anymore.
Nothing to freak out about here.
You could place mics everywhere. The machine will hear everything and know everything unless you whisper