ChatGPT, released in October 2022, was the tool that propelled Generative AI and Large Language Models into the mainstream. It passed the one million userbase mark in five days, setting the record for the fastest-growing software application in history.
Limitations & Risks
Inaccurate, Incomplete Answers
LLMs produce believable answers.
Although believable and assertive, the output from LLMs is not always factual. Due to the Reinforcement Learning (RL) approach with which these models are trained, the models will give out made-up responses from time to time - as long as they are believable. In other words, these models will lie and cheat, behaving like sycophants to make you trust them. Microsoft even has this listed under their FAQs for ChatGPT-based Bing search, advising users to use their judgement and double-check the facts before making critical decisions based on the AI-powered Bing search.