However, when testing the models in a set of scenarios that the authors said were “representative” of real uses of ChatGPT, the intervention appeared less effective, only reducing deception rates by a factor of two. “We do not yet fully understand why a larger reduction was not observed,” wrote the researchers.

Translation: "We have no idea what the fuck we're doing or how any of this shit actually works lol. Also we might be the ones scheming since we have vested interest in making these models sound more advanced than they actually are."