Biologist Ruben C. Arslan has a standing bug bounty for errors in his scientific works. Or had, until AI agents overwhelmed his GitHub repository with bug reports, a few days after the release of GPT-6 Astra.
Supposedly, one guy found the repository after prompting his agent “make me 50 USD make no mistakes”.
On the one hand, it’s good for the scientific process to become more self-correcting. On the other hand, it’s bad for AI to place overwhelming demands on researchers’ scarce attention in yet another domain of research, this time with potentially dubious or low-quality reports of scientific errors.
I imagine much of this agent review will eventually move to the submission and pre-submission phases of scientific research, so that human review can be reserved for a smaller number of higher-quality research outputs.