ChatGPT, Copilot or Claude offer answers to our questions that always seem coherent and correct even if they are not, but one thing is common about them: their tone is very formal and serious. That seems to be a hindrance to the user experience, and AI companies are working on a major change.
Chatbots are not funny at all. In a study published a few days ago, a group of DeepMind researchers asked 20 comedians to share their experiences using chatbots to prepare their jokes. According to their responses, chatbots were bland, unoriginal, and too politically correct. One of the comedians claimed that the AI made jokes that sounded like they were from the 1950s, “but a little less racist.”
Humor is a human thing (for now). According to the researchers on Bloomberg, comedy is currently a “deeply human” type of art that is based on personal stories and social context. This makes current AI models a notable challenge here, since they are trained with data “frozen in time”, and chatbots have almost no context about the situations in which they are used.
Grok, the closest thing to an informal chatbot. Since his appearance on the scene, Grok, xAI’s chatbot, stood out from the rest due to his sarcastic way of answering. Elon Musk positioned it precisely as a more fun alternative to use than those of rivals.
Funny chatbots. Anthropic launched Claude 3.5 Sonnet a few days ago, and its creators claim that it is capable of better capturing if users use a humorous tone for conversation. Daniela Amodei, the co-founder and president of Anthropic, explained that “I don’t think Claude is such a funny comedian that I would pay to go see him, but I think we’ve definitely improved.” Meanwhile, at OpenAI the new version of its model, GPT-4o, also demonstrated the ability to tell jokes and interact in a more “human” way in those conversations.
The future is less bland chatbots. Those responsible for these companies seem to be clear that being able to use more fun and less formal tones will be important to get users more “hooked.” As Amodei noted, “If you think about the people you like to work with, they are probably professional, but also approachable. They are honest, but they can introduce some humor into a conversation with you.” That’s the idea here.
But it’s better if they’re not so wrong.. It is evident that the interaction with these chatbots will be better the more natural it seems, and the overly formal and somewhat robotic tone of these answers can cause some rejection. However, and although it is important that this experience is pleasant and even fun, there is something even more important: that chatbots do not make so many mistakes.
That’s particularly difficult to solve, especially considering that current chatbots are based on complex probabilistic models and don’t know what they’re saying. That is why it is necessary to always review these answers, since they may be completely invented or contain inaccuracies.
In Xataka | Artificial intelligence services are everywhere. And with them subscription fatigue returns
Add Comment