Like many others people last week, Bindu Reddy recently fell under the spell of ChatGPT, a free chatbot that can answer all kinds of questions in an amazing and amazing way.
Reddy, the CEO of Abacus.AI, which makes tools for coders that use artificial intelligence, was impressed by ChatGPT’s ability to respond to requests for love meanings or new recipes for shopping malls. His company is already looking into how to use ChatGPT to help with technical documentation. “We’ve tried it, and it’s working well,” he says.
ChatGPT, developed by the OpenAI startup, has been popular on the Internet since its release last week. Early users quickly posted pictures of their experiments, marveling at its potential Create short stories on each topic, creating literary worksthe answer difficult coding questions, and many more. It has led to predictions that the project will make traditional search engines and home services obsolete.
However the AI at the core of ChatGPT is not, in fact, very new. It is a type of AI called GPT-3 that generates text based on patterns that are analyzed from a large amount of text collected from the Internet. That model, which is available as a commercial API for developers, has already shown that it can answer queries and generate scripts well in some cases. But in order for the project to respond in a certain way, it was necessary to create the right time to feed the program.
ChatGPT is popular because it can take a common question and answer it using a new version of GPT-3, called GPT-3.5. This tweak has opened up new possibilities for answering all kinds of questions, giving the powerful AI model an attractive new look almost anyone can use. That OpenAI has opened up the process for free, and while its distractions can be fun, it also helped start chatbot viruses—just as other AI-based image processing tools have proven useful for creating memes.
OpenAI hasn’t released details about how it gave its text-processing software a new look, but the company did share some details on a blog post. It says the team fed human-written responses to GPT-3.5 as training data, then used a similar reward-and-punishment method known as reinforcement learning to push the model to provide better answers to model questions.
Christopher Potts, a professor at Stanford University, says that the method used to help ChatGPT answer questions, which OpenAI has already demonstrated, is seen as an important step towards helping AI deal with language in a more collaborative way. “It’s very interesting,” Potts says of this approach, although he thinks it could make his job more difficult. “It’s made me think about what I’m going to do in my courses that require short answers in sections,” says Potts.
Jacob Andreas, an assistant professor who works on AI and language at MIT, says the system appears to be expanding the pool of people who can use AI language tools. “This is presented to you in a familiar form that makes you adopt a mental model that you are used to using for other people – people – that you interact with,” he says.