OpenAI has announced GPT-4, the next iteration of its generative AI that its popular ChatGPT is built on. The company says GPT-4 is its “most advanced system, ” which can accept image and text inputs to transmit text outputs. The new multimodal model can be accessed via OpenAI’s ChatGPT Plus $20-a-month subscription, and the API is also available to developers via a waitlist.
“While less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks,” OpenAI announced in a research blog post. Unlike GPT-3.5, which finished around 10 per cent bottom of test takers for a stimulated bar exam, OpenAI said GPT-4 finished around the top 10 per cent. The company said it spent six months testing and training the new language model.
Expanding on capabilities between GPT-3.5 and GPT-4, Open AI says the distinction “can be subtle” in casual conversation. Still, the differences really come out when the tax complexity “reaches a sufficient threshold.” GPT-4 exhibits more creativity and reliability when handling “nuanced instructions” compared to GPT-3.5. Despite this, the company says GPT-4 still has limitations. OpenAI cautions against using the AI model in high-stakes contexts, warning it can “hallucinate” facts and make reasoning errors.
GPT-4 is more advanced than GPT-3.5, accepting the prompt of text and images to output texts. Visually, it has the capability to interpret a wide range of images, including documents with text, photographs, diagrams, or screenshots. Open-AI has started collaborating with companies to integrate GPT-4 into their products, including Duolingo, Stripe, Morgan Stanley and the Government of Iceland. With the power of execution GPT-4 offers, it is no doubt more organisations will be looking to adopt it.
OpenAI says there is still much work to be done on GPT-4, and it looks forward to improving the language model through collective efforts. Microsoft confirmed its new Bing has been running GPT-4, while Google also announced recently it is bringing generative AI into Workspace as the competition heats up.
