The landmark paper that started why there might be doubt when "discovers" are merely mirages produced by small, flexible, biased studies.
GPT-5.1 Pro estimates that 50-70% of all publications are false. Anything above 20 is seemingly alarming until you follow the incentives.