The work raises the obvious question whether this “self-correction” could and should be baked into language models from the start. Wherever that weaker signal originates, the human feedback helps the model boost it when prompted for an unbiased response, says Askell.

“That bias increases with model size.”īut at the same time, somewhere in the training data there must also be some examples of people pushing back against this biased behavior-perhaps in response to unpleasant posts on sites like Reddit or Twitter, for example. “As the models get larger, they also have larger training data sets, and in those data sets there are lots of examples of biased or stereotypical behavior,” says Ganguli. GPT-3 has around 175 billion parameters.) In some cases, the model even started to engage in positive discrimination in its output.Ĭrucially, as with much deep-learning work, the researchers don’t really know exactly why the models are able to do this, although they have some hunches.

(The more parameters, the bigger the model. The team found that just prompting a model to make sure its answers didn’t rely on stereotyping had a dramatically positive effect on its output, particularly in those that had completed enough rounds of RLHF and had more than 22 billion parameters, the variables in an AI system that get tweaked during training.

Tech for less reviews