Meet o1: ChatGPT Creator Releases New AI Model With Better ‘Reasoning’ to Tackle Complex Problems

SAN FRANCISCO, Sept. 13 — ChatGPT creator OpenAI on Thursday released a new series of AI models designed to spend more time thinking — in the hopes that generative AI chatbots will deliver more accurate and beneficial answers.

The new models, known as OpenAI o1-Preview, aim to solve complex tasks and more challenging problems in science, coding, and math — something previous models failed to do consistently.

Unlike their predecessors, these models have been trained to refine their thought processes, try different methods, and recognize errors before providing a final answer.

The new release comes as OpenAI raises funding that could value the company at around $150 billion (650 billion ringgit), which would make it one of the most valuable private companies in the world, according to US media.

Investors include Microsoft and Nvidia, and it may also include a $7 billion investment from MGX, an investment fund backed by the United Arab Emirates, The Information reports.

OpenAI CEO Sam Altman called these models “a new paradigm: AI that can perform complex, general-purpose reasoning.”

But he cautioned that the technology “still has its flaws, is limited, and seems more impressive the first time you use it than after you spend some time with it.”

OpenAI’s drive to improve “thinking” in its model is a response to the persistent problem of “hallucinations” in AI chatbots.

This refers to their tendency to generate persuasive but incorrect content, which has somewhat cooled the enthusiasm of business customers for ChatGPT-style AI features

“We noticed that this model was less likely to hallucinate,” OpenAI researcher Jerry Tworek told The Verge.

But “we can’t say we’ve solved the problem of hallucinations,” he added.

The Microsoft-backed company said that in tests, the models performed comparably better than PhD students solving difficult problems in physics, chemistry and biology.

The students also showed excellent results in mathematics and coding, achieving an 83 percent pass rate on the International Mathematical Olympiad qualifying exam, while GPT-4o, the most advanced general-purpose model, scored only 13 percent.

According to OpenAI, the new inference capabilities could be used by healthcare researchers to annotate cell sequencing data, by physicists to generate complex patterns, and by computer programmers to create and execute multi-step projects.

The company also said that these models have passed rigorous jailbreaking tests and are more resistant to bypassing security attempts.

OpenAI said its enhanced security measures also include recent agreements with the US and UK AI Security Institutes, which were granted early access to models for evaluation and testing. — AFP