AI chatbots are getting weirder and some observers say the problem could lie in the large language models (LLMs) used to create them.
“What's Wrong with LLMs and What Should We Build Instead” – Tom Dietterich – #VSCF2023
Microsoft’s AI Bing chatbot generates strange or aggressive responses to questions. The software generates text based on prompts and has sparked rumors that chatbots could become sentient. That’s not true, said Evan Coopersmith, executive vice president of data science for software company AE Studio.
"With large language models, the training process is literally a subset of what's known as reinforcement learning," Coopersmith told Lifewire in an email interview. "Just like parenting, we're reinforcing what we believe is the right behavior."
When Microsoft recently launched its Bing chatbot, the company apparently expected its creation to generate relatively mundane answers to common internet queries. Things began to go awry when users engaged in extended conversations with the chatbot, which is powered by a large language model (LLM), a computer program for natural language processing.
