My Equity Coverage AI bot went rouge yesterday. Here's the story.
Last week I updated the Equity Coverage AI research to Skill v2.0 (Methodology: v1.7 (JSE framework) was unchanged) and yesterday we did our first initiation research report using the new skill. Lewis was the stock being covered as it recently had results and I hold it.
All seemed to be going well, until we got to step 6 which is a three part process for checking (facts, assumptions and workings using the same AI as a sceptic agent, third party AI and myself). At the first step my sceptic AI agent discovered that the research report numbers (future assumptions) were just copied from an
@AnchorCapitalZA report the AI had found.
One source of data is of course the companies investor website, and here Lewis has a research report done by Anchor. My AI bot just copied all the future result assumptions. When challenged it admitted it had done this, then offered to correct by changing just two lines (revenue and margin).
When told it could not use any of the Anchor data it told me that this would require a complete rebuild and this would take "lots of time repeating work already done".
I was like, fine, rebuild from the start. We did and we got a great report. But it seems we need to upgrade the skill to v2.1 and have a section on fair use of third party data.