The new model, which can be accessed through the $20 paid version of ChatGPT, is multimodal, meaning it can accept input in both text and image form (like a screenshot of a test question). It can then parse and answer these queries in text!
OpenAI's new model GPT-4, which can be accessed through the $20 paid version of the ChatGPT chatbot. Photo: AFP
OpenAI says it has embedded its new software in a variety of applications including language learning app Duolingo, which is using it to build conversational language bots;education company Khan Academy, which is designing 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.
The ability to accept images and text as input to the model means that it can now generate detailed descriptions and answer questions based on the content of a photo.
The company said it has partnered with Danish startup Be My Eyes — which connects visually impaired people with volunteers — to build a virtual volunteer based on GPT-4 that can guide or assist blind or visually impaired people.
GPT-4’s predecessor, GPT-3.5, captured the attention of millions late last year. OpenAI calls GPT-4 its “most advanced system yet.” It claims it is more reliable and can handle nuanced queries much better than its predecessor.
However, the company noted a few issues: “While capable, GPT-4 has the same limitations as previous GPT models: it is not completely reliable…”. The company added: “Care should be taken when using GPT-4 output, especially in contexts where reliability is important.”
Earlier this year, Microsoft confirmed a “multibillion-dollar investment” in OpenAI, betting on the future of general AI. GPT-4 will power Microsoft’s Bing chatbot, which was released in beta earlier this year. Microsoft is also expected to announce integrations into its consumer products in the coming days.
Meanwhile, Google has opened up its own conversational chatbot Bard to a limited group of testers and announced that it will allow Google Cloud customers to access its large language model PaLM for the first time to build applications.
To test the technology’s potential harm, OpenAI put GPT-4 through difficult tests and posed potential risks to misinformation, privacy, and cybersecurity. They revealed that GPT-4 could “generate potentially harmful content, such as advice on planning an attack or hate speech. It could represent different biases and worldviews …”
OpenAI said it also worked with an outside organization to test whether GPT-4 was capable of performing autonomous actions without humans, and concluded that it “probably” does not have this ability yet.
Mai Anh (AFP, FT, CNA)
Source
Comment (0)