American artificial intelligence (AI) research organization, OpenAI has unveiled its much anticipated GPT-01 model which has scored impressively on benchmarks.
The company released a sneak peek of its next-generation large language models on Friday. It claims that these models outperform earlier models, but some limitations exist.
When OpenAI unveiled the o1-preview model, it boasted about how well the model performed on a range of human-designed tasks: in Codeforces programming competitions, the model achieved an 89th percentile score, and on an International Mathematics Olympiad qualifying test, it answered 83 percent of the questions correctly, compared to 14 percent for GPT-4o.
READ ALSO: Microsoft and Open AI face legal action over alleged copyright infringement
‘‘The o1-preview and o1-mini models represent the “beginning of a new paradigm: AI that can do general-purpose complex reasoning,” according to OpenAI CEO Sam Altman. Additionally, he stated that “o1 is still limited, has flaws, and seems more impressive when you first use it than it does when you use it more often.”
Additionally, according to OpenAI, intentional hallucinations generally occur when o1-preview is requested to supply references to papers, websites, books, or other comparable sources that are difficult for it to independently check without access to the internet.
“Intentional hallucinations primarily happen when o1-preview is asked to provide references to articles, websites, books, or similar sources that it cannot easily verify without access to internet search, causing o1-preview to make up plausible examples instead,” the company wrote in its model system card.
The post OpenAI unveils new GPT-01 model…but with limitations appeared first on Latest Nigeria News | Top Stories from Naomisophyblog.
