The Microsoft funded start-up OpenAI has released its next generation AI chatbot – meet the more powerful and, it is claimed, even more reliable ChatGPT-4.
The upgraded “multimodal” ChatGPT-4 (GPT-4) “can solve difficult problems with greater accuracy, thanks to its broader general knowledge and problem solving abilities.” according to OpenAI.
One unique feature is the chatbot’s ability to generate illustrative text and images based on any input prompt from the user.
According to developers, what makes this new AI model superior over its predecessor is its “advanced reasoning capabilities” which produce “safer and more useful responses.”
Developers said they took six months to scale up deep learning to ensure the fourth iteration of ChatGPT would be safer and more aligned.
To improve the chatbots’ overall behavior, developers said they’ve utilized feedback from over 50 experts in areas such as AI safety and security, as well as feedback submitted from ChatGPT users.
“GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5 on our internal evaluations,” OpenAI claimed.
Released in November, ChatGPT has been observed by users producing, at times, disturbing content on ethical questions about humanity, as well as AI hallucinations, in which the chatbot generates completely false content but presents it as fact.
“This technology will profoundly transform how we live. There is still time to guide its trajectory, limit abuse, and secure the most broadly beneficial outcomes,” said Anna Makanju, in charge of public policy at OpenAI.
GPT-4 can generate, edit, and iterate with users on creative and technical writing tasks, such as composing songs, writing screenplays, or learning a user’s writing style,” the company said.
According to OpenAI internal statistics, GPT-4 easily outscores ChatGPT on simulated exams.
OpenAI president and co-founder Greg Brockman showcased some of GPT-4's abilities during a developer demo Tuesday, which was praised by those who tuned into the recorded live stream.
“I'm impressed how GPT-4 can read doctor's handwriting,” one viewer commented.
“GPT-4's applications are vast, ranging from creating training materials to performing mathematical calculations. The technology has the potential to transform many industries, as it understands user intent from context and can even analyze images,” wrote another viewer.
“Amazing demo. The napkin-to-HTML one was mind-blowing. There are so many applications, you could make glasses that give descriptive audio to the visually impaired and so on. The future is nuts, “ commented another.
The welcome reception is a far cry from Google’s attempt to introduce its own AI chatbot competition known as Bard.
During Google’s live YouTube presentation – designed to show off Bard’s capabilities – the chatbot instead returned wrong answers to input questions, and moments later the stream abruptly was taken down.
The February flop of its ChatGPT rival cost the company billions in value.
Still, according to OpenAI, GPT-4 still has many known limitations to address, including social biases, hallucinations, and adversarial prompts.
Currently, GPT-4 is available on ChatGPT Plus, a monthly paid subscription for users.
OpenAI is has also given access to API for developers to build applications and services.
Microsoft, which has poured millions of dollars into OpenAI since the ChatGPT release, announced the GPT-4 version has now been formally integrated into its current Bing browser.
The software giant will be incorporating ChatGPT-4 into its Azure Cloud and other Microsoft technology.
The move is expected to create a financial windfall and guaranteed place on top of the AI chatbot competition for Microsoft and OpenAI.
“AI systems are becoming a part of everyday life. The key is to ensure that these machines are aligned with human intentions and values,” said Mira Murati, chief technology officer at OpenAI.
More from Cybernews:
Subscribe to our newsletter