OpenAI released a new generation of large multi-modal model GPT-4
OpenAI has released a new generation of large-scale multimodal model, GPT-4, which can accept text and image inputs. GPT-4 has achieved human-level performance on various professional and academic benchmarks. For example, in simulating the lawyer exam, GPT-4 scored in the top 10%, while GPT-3.5 scored in the bottom 10%.
In multilingual testing, GPT-4 outperformed GPT-3.5 and Chinchilla (from DeepMind) and PaLM (from Google), including English performance, as well as low-resource languages such as Latvian, Welsh, and Swahili.
OpenAI has also opened up the ability for role-playing and personality customization, allowing developers and users to customize their AI style rather than having the classic ChatGPT personality with fixed length, tone, and style.
ChatGPT Plus subscription users can now directly use GPT-4, and a certain number of GPT-4 experiences will be opened to free users in the future. The GPT-4 API requires application to the waiting list, with a price of $0.03 per 1k prompt tokens and $0.06 per 1k completion tokens. Currently, image input is in the research preview phase and is only available to a few customers.
Microsoft has also officially confirmed that Bing Chat is running on GPT-4, and the Edge sidebar feature of Bing Chat has been launched. Unlike the offline version of GPT-4 with data stopping in September 2021, Bing Chat can obtain real-time information online and is free to use.
