Hey folks, I just received an email from Humata about an update to their terms regarding AI training. The gist is that they're formalizing the process for opting out of having your data used to train their models.
I've always appreciated their transparency on data usage, but this shift to a more explicit opt-out system has me wondering about the practical implications. Will opting out affect the performance of the AI on my own documents? How streamlined will the opt-out mechanism be in the dashboard?
I think it's a positive move towards user control, but the details matter. Has anyone else looked into this yet or have thoughts on how this compares to other platforms' approaches? Let's discuss the specifics and any potential workflow considerations.
Keep it civil, keep it real.
I received the same email and have been dissecting the specifics. You raise a critical question about performance impact on your own documents. Based on the language used, my reading is that opting out would only preclude your data from being used in the general, aggregate model training datasets. The inference engine used for your personal queries, which operates on your uploaded documents in a more isolated context, should theoretically remain unchanged. However, the absence of a clear architectural diagram in their communication makes this an assumption.
Regarding the dashboard mechanism, I've compared this to the opt-out workflows of three other major platforms I audit. The trend is toward a single, account-level toggle, which is clean but often lacks granularity (e.g., per-document or per-project controls). If Humata's implementation follows that pattern, it will be streamlined for user control but potentially problematic for teams managing varied document sensitivities within a single account. The real workflow consideration will be whether the setting is retroactive for previously uploaded documents or only applies to new uploads post-opt-out; the email was ambiguous on that point.