OpenAI has released GPT-4, its latest artificial intelligence model that it says demonstrates “human-level performance” on several academic and professional benchmarks, such as the US Bar Exam, Advanced Placement Tests, and the School SAT exams.
The new model, which is accessible through the $20 paid version of ChatGPT, is multimodal, meaning it can accept input in both text and image form. It can then analyze and answer these questions using text.
OpenAI said it has embedded its new software in a variety of apps, including language learning app Duolingo, which it uses to build conversational language bots; education company Khan Academy, which has designed an online tutor; and Morgan Stanley Wealth Management, which is testing an internal chatbot that uses GPT-4 to retrieve and summarize information for its employees.
With the ability to accept images and text as input, the model can now generate detailed descriptions and answer questions based on the content of a photo. The company said it is teaming up with Danish start-up Be My Eyes — which connects visually impaired people with human volunteers — to build a GPT-4-based virtual volunteer that can guide or assist the blind or visually impaired.
GPT-4’s predecessor, GPT-3.5, captured the imagination of millions of people late last year who experimented with the question-and-answer chatbot ChatGPT.
According to OpenAI, GPT-4 is the “most advanced system yet”. It claims to be more reliable and much more capable of handling nuanced queries than its predecessor. For example, GPT-4 scored in the 90th percentile on the Uniform Bar Exam administered by aspiring lawyers in the US compared to ChatGPT, which only reached the 10th percentile.
However, the company noted some issues: “Despite its capabilities, GPT-4 has similar limitations to previous GPT models: it is not completely reliable (for example, may suffer from “hallucinations”), has a limited context window, and does not learn from experience. ”
“Caution should be exercised when using the output of GPT-4, especially in contexts where reliability is important,” the company added.
Earlier this year, Microsoft confirmed a “multi-billion dollar investment” in OpenAI over several years, placing a bet on the future of generative AI – software that can respond to complex human queries in natural-sounding language. GPT-4 will support Microsoft’s Bing chatbot, which was released in limited edition earlier this year. Microsoft is also expected to announce the integration into its consumer products in the coming days.
Meanwhile, Google has opened up its own conversational chatbot, Bard, to a limited pool of testers and announced that it will give Google Cloud customers access to its large language model PaLM to build applications for the first time.
OpenAI, which had published some details of previous models such as GPT-3, said it would not reveal details about the technical aspects of GPT-4, including the architecture of the model, what data it was trained on, or the hardware and computers. capacity used to deploy it, due to competition and security concerns.
To test the damage of the technology, the company stress-tested GPT-4 and outlined the risks it foresees around bias, disinformation, privacy and cybersecurity. They revealed that GPT-4 “may generate potentially harmful content, such as advice on attack planning or hate speech. It can represent different prejudices and worldviews. . . it can also generate code that is compromised or vulnerable.” They said it can provide detailed information about carrying out illegal activities, including developing biological weapons.
OpenAI also said it was working with a third-party organization to test whether GPT-4 was capable of performing autonomous actions without human input and concluded that it was “probably” not yet capable of doing so.
Additional reporting from Richard Waters