According to TapTechNews on June 22, as reported by TechXpore yesterday, an AI research team of Google's DeepMind project has recently found that large language models are not very good at writing interesting jokes.
In a new attempt, the team tried to understand whether large language models can create jokes or even a complete stand-up comedy. For this purpose, the team recruited 20 professional stand-up comedians who have used large models in their work before. These comedians need to use large models to help them create a complete stand-up comedy and rate the results.
The researchers found that large language models are quite good at making jokes, but only a small portion of them are really funny. They believe that most of the jokes generated are vague and there are few surprising results.
Many comedians think that the jokes generated by AI lack sharpness, which is one of the factors needed to make jokes interesting. Among them, many also think the results are plain and unremarkable. But there are also some comedians who believe that large language models are useful in generating routines, and they can create jokes around these routines.
The research team believes that considering that the makers of large language models will use filters to prevent output that may be considered offensive or offensive, it is not surprising that the jokes told by AI are not funny enough.
TapTechNews attaches the paper address: https://arxiv.org/abs/2405.20956