Source: AFP
ChatGPT creator OpenAI on Thursday released a new line of AI models designed to spend more time thinking — with the hope that the resulting AI chatbots will provide more accurate and helpful answers.
The new models, known as OpenAI o1-Preview, are designed to tackle complex tasks and solve more difficult problems in science, coding and math — something that previous models have been criticized for not delivering consistently.
Unlike their predecessors, these models are trained to refine their thought processes, try different methods and recognize mistakes before developing a final answer.
The new release comes as OpenAI raises funds that could see it valued at around $150 billion, which would make it one of the most valuable private companies in the world, according to US media.
Read also
Ireland launches EU privacy probe into Google AI development
Investors include Microsoft and Nvidia and could also include a $7 billion investment from MGX, an investment fund backed by the United Arab Emirates, The Information reported.
OpenAI CEO Sam Altman hailed the models as “a new paradigm: AI that can do complex, general-purpose reasoning.”
However, he cautioned that the technology “is still flawed, limited, and still seems more impressive on first use than when you spend more time with it.”
OpenAI’s push to improve the “thinking” in its model is a response to the persistent problem of “hallucinations” in AI chatbots.
This refers to their tendency to create persuasive but flawed content which has somewhat cooled enthusiasm for ChatGPT-style AI features among business customers
“We’ve noticed that this model has fewer hallucinations,” OpenAI researcher Jerry Tworek told The Verge.
Read also
Sony will release the PlayStation 5 Pro on November 7th
But “we can’t say we solved hallucinations,” he added.
The Microsoft-backed company said that in tests, the models performed comparable to doctoral students on difficult tasks in physics, chemistry and biology.
They also excelled at math and coding, achieving an 83 percent pass rate on qualifying exams for the International Mathematical Olympiad, compared to 13 percent for the GPT-4o, its most advanced general-purpose model.
OpenAI said the new reasoning capabilities could be used by healthcare researchers to annotate cell sequence data, physicists to create complex formulas, or computer programmers to build and execute multi-step designs.
The company also said the models survived rigorous jailbreaking tests and could better withstand attempts to bypass its protective bars.
OpenAI said its enhanced security measures also included recent agreements with the US and UK AI Security Institutes, which were granted early access to the models for evaluation and testing.
Source: AFP
