Most Robust Interventions?

What do you think the most robustly good charitable interventions are?

I don’t mind if the intervention has a good chance of not being effective after all, but I am worried underestimated consequences might turn the EV of charities negative.

Background: I think many attempts to do good have a considerable risk of being net-negative, including things popular in EA.

For example, 80k motivating EAs to work at AI companies might have been a mistake. Saving human lives might increase meat consumption. Informing governments about AGI might motivate them to race.

I am not claiming these backfire effects are real or that they outweigh the upsides, but I am not comfortable ruling it out either.

I currently tend to think humane slaughter initiatives like the Shrimp Welfare Project are rather unlikely to backfire, because they don’t change the number of individuals, and the affected individuals don’t majorly affect the world after the intervention, because they die anyway. It might backfire if the stunning mechanism or similar malfunctions, making it worse than default slaughter, but this becomes less likely the more suffering the default slaughter involves already, like drawn-out slow suffocation.