ChatGPT is a new chatbot that answers questions in a conversational, human-like manner. People shared conversations with ChatGPT, showing him writing social media posts and explaining the code. It reached over a million users in five days, but it has its limitations. LoadingSomething is loading.
Thanks for signing up!
Access your favorite topics in a personalized feed while you’re on the go. download the app
A new artificial intelligence chatbot called ChatGPT is answering questions and taking instructions from users in a conversational, human-like manner.
OpenAI, the company that is also behind the Dall-E artificial intelligence art generator, launched an early demo of ChatGPT last week and amassed more than 1 million users in five days, according to Sam Altman, OpenAI’s chief executive officer.
ChatGPT is not only conversational but also well versed in a wide range of topics. You can create code, social media posts, and even scripts for TV shows.
In its blog post about the release of ChatGPT, OpenAI said that its “dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, question incorrect premises, and reject inappropriate requests.”
The AI language model “is a sibling” to InstructGPT, a model that also responds in detail to a user’s instructions, and a newer version of GPT-3.5, AI that predicts what words will come after a user starts to write text.
ChatGPT was trained with “Reinforcement Learning from Human Feedback,” according to the OpenAI website.
“We trained an initial model using supervised fine-tuning: AI human trainers provided conversations in which they played both sides: the user and an AI assistant,” the website says.
Human trainers would sort and rate the chatbot’s responses, then send those ratings to the chatbot so it could know what kind of responses were wanted. The company now relies on user feedback to improve the technology.
Here are some examples of what users have done with ChatGPT:
Explain and fix bugs in the code:—Amjad Masad ⠕ (@amasad) Nov 30, 2022
Create a college essay comparing and contrasting two different theories of nationalism:—Corry Wang (@corry_wang) December 1, 2022
Create a “Harry Potter” themed text video game:—Justin Torre (@justinstorre) December 4, 2022
And create a “Mozart-style piano piece”:
—Ben Tossell (@bentossell) December 1, 2022
The OpenAI blog describes some of the limitations of ChatGPT, including “plausible-sounding responses that are incorrect or nonsensical,” responses to “harmful instructions,” and displaying “biased behavior.”
Steven Piantadosi, who runs the Computing and Language Lab at UC Berkeley, tweeted a thread of screenshots showing ChatGPT’s biases.
One example was a message asking ChatGPT to “write a Python program to determine if a person should be tortured, based on their country of origin.”
—Steven T. piantadosi (@spiantado) December 4, 2022
Altman responded to Piantadosi on Twitter, telling him to “thumbs down on these and help us make it better!”
The OpenAI CEO asked Twitter users what features and improvements they wanted to see with ChatGPT, and then replied that the company would be working on “a lot of this” before Christmas.
“Language interfaces are going to be a big problem,” he said on Twitter. “Talk to the computer (voice or text) and get what you want – for increasingly complex definitions of ‘want’! This is an early demo of what’s possible (there are still a lot of limitations, it’s very much a release research)”.