OpenAI Launches GPT-4, a Multimodal AI Model with State-of-the-Art Capabilities
OpenAI unveils GPT-4, a highly advanced AI that processes both text and images while achieving human-level scores on professional exams. The new model is currently available to paying subscribers and powers Microsoft's Bing Chat.
OpenAI releases GPT-4, a powerful new artificial intelligence model that processes both text and image inputs to deliver state-of-the-art performance. The company makes this upgraded system available immediately to ChatGPT Plus subscribers, while developers join a waitlist to access the model through a priced API. Notably, Microsoft confirms that its Bing Chat chatbot already runs on this new underlying technology.
The latest model demonstrates significant cognitive leaps over its predecessor, GPT-3.5, by achieving human-level results on various academic and professional benchmarks. For instance, GPT-4 scores in the top 10% on a simulated bar exam, a massive improvement over the previous model's bottom 10% score. OpenAI spends six months refining the system through adversarial testing to improve its factuality, steerability, and adherence to safety guardrails.
Major companies quickly integrate GPT-4 into their own services to leverage its enhanced capabilities. Stripe uses the AI to summarize business websites for support staff, Duolingo builds it into a new language learning tier, Morgan Stanley employs it to retrieve company documents for analysts, and Khan Academy creates an automated tutoring system. Additionally, a new feature uses GPT-4 to act as a virtual volunteer that describes the visual world for people who are visually impaired.