OpenAI announces ChatGPT successor GPT-4
The new model can respond to images - for example, it can offer recipes based on photos of items, write captions and descriptions.
It can handle 25,000 words, eight times more than ChatGPT.
Millions of people have been using ChatGPT since its launch in November 2022.
Popular requests for it include songwriting, poetry, promotional copy, computer coding, and homework help, although teachers say students shouldn't use it.
ChatGPT answers questions in human-like natural language, and in 2021 it will use the Internet as a knowledge base to reflect other forms of writing, such as songwriters and writers.
There is concern that one day they will take on many of the jobs people do today.
OpenAI says it spent six months working on GPT-4's security features, training it based on human feedback. However, it warns that they may still be prone to leaking false information.
GPT-4 will first be available to ChatGPT Plus subscribers who pay $20 per month for premium access to the service.
It currently powers Microsoft's Bing search engine platform. The tech giant has invested $10 billion in OpenAI.
The live demo provided answers to complex tax questions, but there was no way to validate the answers.
Like ChatGPT, GPT-4 is a form of generative artificial intelligence. Generative AI uses algorithms and text prediction to create new content based on your requests.
 |
| GPT-4 generates recipes from an image |
According to OpenAI, GPT-4 has "improved reasoning capabilities" compared to ChatGPT. For example, a model that finds available meeting times for three schedules.
OpenAI also announced new partnerships with language learning app Duolingo and Be My Eyes, an app for the blind that builds AI chatbots.
However, OpenAI warned that, like its predecessors, GPT-4 is still not completely reliable and can cause "hallucinations" - a phenomenon in which AI misrepresents facts or makes logical errors.