March 15, 2023
OpenAI has released GPT-4, the most recent and advanced iteration of its text generation technology. GPT-4’s enhanced capabilities are poised to revolutionize AI products from not only OpenAI, but also Microsoft and other industry players.
This ground-breaking launch has the potential to usher in a new era of natural language comprehension and processing, paving the way for more complex and nuanced digital interactions. OpenAI claims that GPT-4 outperforms ChatGPT’s advanced reasoning capabilities, with scores in the 90 percent range for bar exams taken by lawyers, compared to ChatGPT’s 10 percent.
OpenAI’s paying users can now access GPT-4 through ChatGPT Plus, but with a usage limit. Developers interested in using the API can sign up for a waitlist. The cost of using GPT-4 is $0.03 for every 1,000 “prompt” tokens (equivalent to around 750 words) and $0.06 for every 1,000 “completion” tokens (also about 750 words). Tokens refer to raw text representations, where each word is split into smaller segments. “Prompt” tokens are word segments that are entered into GPT-4, whereas “completion” tokens are GPT-4 generated content.
When given an image, the latest version of GPT-4 can now process longer entries of up to 25,000 words and generate captions and other information. OpenAI used human feedback during the testing process to improve the safety and naturalness of GPT-4 over previous versions. When compared to GPT-3.5, GPT-4 is 82% less likely to provide inappropriate content and 40% more likely to generate accurate responses in internal testing.
Following the release of GPT-4, Microsoft announced that its new Bing search engine would use GPT-4 that had been customized for search purposes. For the past five weeks, it has been available on Bing preview.
The sources for this piece include articles in Axios and TechCrunch.
