Today I was (once again) seeing instances where a bot’s response would fail to reflect information that is clearly defined in one of its dozen knowledge base files. When this occurred, the “Message Insights” and “Explore” screens would both show that relevant files were found, but the chunks shown as relevant on the first screen were different than the ones on the second.
Most of the files in this KB are 1 MB or less, but one is nearly 2 MB. On a whim, I broke the 2 MB file into two files. Repeated instances of the same query are showing that responses and chunk listings are correct when I use the smaller files, and they’re frequently incorrect when I switch back to the original large file.
In fact, if I switch on both the two smaller files AND the original large file, the queries also usually fail. The mere fact that the large file is connected appears to cause most queries to fail. (The two smaller files hold the exact same content as the single big one, so this must be a size issue, not a content issue.) When I do occasionally have a correct response with the large file active, the chunks on the “Message Insights” appear to be the relevant ones and the wholly different chunks on the “Explore” screen do NOT contain relevant information.
Given that I believe the stated maximum allowable file size is 100 MB, I would not expect an issue.
Anyone else seeing similar behavior?