Tweeted By @GaryMarcus
New evidence from @Google that size isn't everything: "While model scaling alone can improve quality, it shows less improvements on safety and factual grounding"
— Gary Marcus (@GaryMarcus) January 21, 2022
Translation: making GPT-3-like models bigger makes them more fluent, but no more trustworthy. https://t.co/BUjQCP6ReZ