(Summarized by GPT-4 from: https://techcrunch.com/2023/03/14/with-evals-openai-hopes-to-crowdsource-ai-model-testing/)
OpenAI has introduced GPT-4, a new AI model capable of understanding both text and images, surpassing its predecessor GPT-3.5. GPT-4 is available to paying users via ChatGPT Plus and developers can join a waitlist to access the API. Microsoft’s Bing Chat, Stripe, Duolingo, Morgan Stanley, and Khan Academy are among the early adopters of GPT-4.
This new model performs at a “human level” on professional and academic benchmarks, showing improvements in factuality, steerability, and adherence to boundaries. It can generate text, caption images, and interpret complex visuals. OpenAI has partnered with Be My Eyes for initial testing of GPT-4’s image understanding capabilities.
GPT-4 also introduces “system” messages, allowing developers to prescribe style and task-specific directions. Despite these upgrades, GPT-4 can still “hallucinate” facts and make reasoning errors. However, it has seen improvements in refusing harmful requests and sensitive content compared to GPT-3.5. OpenAI is optimistic about GPT-4’s potential to improve people’s lives through various applications.
Leave a Reply