5la9777kd
@5la9777kd
OpenAI's new paper explains how even a small amount of bad training can mess up AI models. But the good news? Fixing it is usually simple. Earlier this year, researchers found that tweaking GPT-4 with flawed code could cause issues. Still, the study shows these problems aren’t hard to solve.
0 reply
0 recast
0 reaction
dfdh
@sdovosdn
It's awesome that OpenAI found fixing AI glitches is often a breeze even when small training mistakes happen
0 reply
0 recast
0 reaction