OpenAI has just announced the release of its latest language model, GPT-4, which is more creative and collaborative than its predecessors. GPT-4 is capable of generating, editing, and iterating on creative and technical writing tasks such as composing songs, writing screenplays, and learning a user’s writing style. It can also accept images as inputs and generate captions, classifications, and analyses.
One of the most significant improvements in GPT-4 is its ability to handle over 25,000 words of text, allowing for use cases such as long-form content creation, extended conversations, and document search and analysis. Additionally, GPT-4 has outperformed its predecessor, ChatGPT, in several benchmarks, including the Uniform Bar Exam and Biology Olympiad with scores improving from 10th and 31st percentile to 90th and 99th percentile. This is a mammothian improvement.
OpenAI has focused on making GPT-4 safer and more aligned with its goal of creating increasingly sophisticated and capable language models. They have incorporated more human feedback, including feedback from ChatGPT users, and worked with over 50 experts for early feedback in domains such as AI safety and security. OpenAI also plans to continuously improve GPT-4’s safety research and monitoring system.
However, GPT-4 still has known limitations that OpenAI is working to address, such as social biases, hallucinations, and adversarial prompts. OpenAI encourages transparency, user education, and wider AI literacy as society adopts these models. They also aim to expand the avenues of input people have in shaping their models.
GPT-4 was trained on Microsoft Azure AI supercomputers, and its AI-optimized infrastructure allows OpenAI to deliver GPT-4 to users worldwide. GPT-4 is now available on ChatGPT Plus and as an API for developers to build applications and services. Overall, GPT-4 is a significant improvement over its predecessors, and its release represents another step forward in the development of sophisticated language models.
You can catch the developer demo livestream here showing GPT-4 and its capabilities/limitations, which is just an hour away.
