OpenAI Unveils GPT-4, a More Accurate and Creative Language Model with Multimodal Input and Output Capabilities

According to a report from Bloomberg on March 15th, OpenAI announced the latest version of its large language model - GPT-4, which is expected to be more accurate, creative, and collaborative. This news is particularly relevant for electronic component distributors because GPT-4 is a powerful tool that can handle text and image inputs, and it will be used in the paid version of ChatGPT Plus.

The new model is said to be "larger" than previous versions and is able to process text and images to produce output that is "more creative and collaborative than ever before." OpenAI claims that GPT-4 has a wider range of common sense and problem-solving abilities, which enables it to more accurately solve complex problems. In internal evaluations, it has shown to be 40% more likely to produce correct responses than GPT-3.5.

GPT-4 has been designed as a multi-modal tool that supports both text and image input features. It generates fewer incorrect answers, deviates less from the conversation path, and avoids taboo topics better than previous versions. In fact, in many standardized tests, GPT-4 outperformed humans. For example, it ranked in the top 10% of test-takers for a simulated lawyer exam, in the top 7% for SAT reading, and in the top 11% for SAT math.

However, OpenAI has also warned that GPT-4 is not perfect, and in many cases, its abilities are not as good as humans. The company is working to address known limitations, such as social biases, illusions, and adversarial prompts.

According to the report, Microsoft has stated that the new artificial intelligence tool is being used to support its Bing search engine. Additionally, Morgan Stanley is using GPT-4 to organize data, and the electronic payment company Stripe is testing whether it can help combat fraud.