ChatGPT's breakthrough is not what it seems.
The model learns that hedging is a signal of lower-quality output. This creates a systematic bias toward sounding certain.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results