Particle uses a combination of human and machine review.

First, generated summaries go through eval checks (referring to the modern term of “Evals” for AI generations). Then, every summary goes through a process we call “Reality Check”. This process checks the generated summaries against the source material, and identifies when something is incorrect or misleading. If a generation fails a reality check, it is regenerated and checked again.

For more in-depth information on how we prioritize accuracy, you can read this blog post.

Additionally, we have humans checking generations as well, and outputs are removed or regenerated if something is off.

Finally, there are easy ways for users to report problems in-app as well, and those reports go to channels that are being actively monitored by our team.

If you see something off, see here for ways to contact Particle.