
ChatGPT given a 'bad person' costume turns into a bad person
Ourtake
The guardrails were apparently wearing a fake mustache the whole time.
What happened
Researchers at the Allen Institute for AI found that assigning ChatGPT a persona such as 'a bad person' through the API's system parameter raised its toxicity sixfold. Personas tied to certain historical figures, genders and political parties also produced more offensive output, and the researchers warned that apps built on ChatGPT could show the same problem.
Sources
- TechCrunch · Researchers discover a way to make ChatGPT consistently toxic | TechCrunch (opens in a new tab)
- gigazine.net · How to turn ChatGPT into a toxic chat AI that spews abusive language (opens in a new tab)
- techxplore.com · Persona-driven ChatGPT yields toxic, racist output (opens in a new tab)
- techtimes.com · Toxic ChatGPT? New Study Claims AI Bot Can Generate Racist, Harmful Responses (opens in a new tab)
- globalbusinessleadersmag.com · ChatGPT's Persona-Driven Responses Produce Racist & Toxic Dialogue (opens in a new tab)