OpenAI ChatGPT releases GPT-4; new AI model
OpenAI announced on Tuesday it is releasing a new version of its widely known artificial intelligence software ChatGPT, named GPT-4.
On its official website, OpenAI described the new model to possess “broader general knowledge and problem-solving abilities.” This enables it to “solve difficult problems with greater accuracy,” the company said.
The website specified the new version is more responsive to creative and technical writing tasks; including songs composing and screenplays writing. The new model is also capable of handling long texts exceeding 25,000 words.
Safety & Biases
OpenAI’s CEO, Sam Altman described the new model to be more creative and less biased than its predecessor. With regards to safety, the developing company mentioned it has “incorporated more human feedback… to improve GPT-4’s behaviour.”
GPT-4 is 82 percent less likely to respond to requests for disallowed content when compared to ChatGPT.
Sarah Myers West, managing director of the AI Now Institute which studies the effects of AI on society, expressed how monitoring is needed to accompany the release of such tools.
“We have clear evidence that generative AI systems routinely produce error-prone, derogatory and discriminatory results,” West told NBC.
OpenAI said it is cooperating with a number of organisations to incorporate GPT-4 in building innovative projects. Such organisations include Duolingo, Be My Eyes, in addition to Morgan Stanley.
Bill Peterson, lead engineer at Duolingo said “GPT-4 gives us far more confidence in the accuracy of AI responses in Explain my Answer,” referring to a feature which teaches the language-learning app users grammar rules.
In his tweets, Altman asserted the model is “still flawed, still limited,” therefore him and his company would “appreciate feedback on its shortcomings.”
OpenAI follows the principles of open-source software, which gives the opportunity for developers to contribute to the original model released by the company.
Announcing GPT-4, a large multimodal model, with our best-ever results on capabilities and alignment: https://t.co/TwLFssyALF pic.twitter.com/lYWwPjZbSg
— OpenAI (@OpenAI) March 14, 2023
we have had the initial training of GPT-4 done for quite awhile, but it’s taken us a long time and a lot of work to feel ready to release it. we hope you enjoy it and we really appreciate feedback on its shortcomings.
— Sam Altman (@sama) March 14, 2023