A lot of vendors and open-source projects are dealing with a pretty serious problem lately: a flood of AI-generated “bug reports”. These reports are, put politely, confident garbage.

It reminds me of the prize offered for solving Fermat's Last Theorem and the experience the Wolfskehl committee had. Quoting Wikipedia:

“Prior to Wiles's proof, thousands of incorrect proofs were submitted to the Wolfskehl committee, amounting to roughly 10 feet (3.0 meters) of correspondence. In the first year alone (1907–1908), 621 attempted proofs were submitted, although by the 1970s, the rate of submission had decreased to roughly 3–4 attempted proofs per month. According to some claims, Edmund Landau tended to use a special preprinted form for such proofs, where the location of the first mistake was left blank to be filled by one of his graduate students”

The real problem now is that a lot of these AI-written so called “bug reports” sound just plausible enough that you can’t automatically dismiss them. So, instead of building trust between the external community and engineering teams, companies end up wasting expensive resources reviewing mountains of garbage.