OpenAI unveiled its latest o1 models on Thursday, which offer significant advancements in logical thinking, problem-solving, and answering scientific questions. The models are seen as a crucial step towards the development of artificial general intelligence (AGI) – machines that achieve human-like cognitive abilities.
In the accompanying system card describing the functionality of the AI, OpenAI acknowledges, however, that the new models have also "significantly increased" the risk of technology misuse, particularly in connection with the creation of biological weapons. For the first time, the company assigned a "medium risk rating" to its models concerning chemical, biological, radiological, and nuclear (CBRN) weapons. This implies that the technology has significantly enhanced experts' abilities to develop biological weapons.
Advanced AI software capable of performing step-by-step logical reasoning carries an increased potential for misuse, according to experts. Yoshua Bengio, one of the world's leading AI researchers, emphasized the urgency of legal regulations to mitigate these risks. He pointed to a legislative bill in California that would require companies to implement safety measures to prevent the use of AI in the development of bioweapons. "The more AI advances towards AGI, the greater the risks if appropriate safeguards are not in place," warned Bengio.
Technology companies like Google, Meta, and Anthropic are also currently working on developing advanced AI systems. These AI agents could assist people in completing tasks and offer potentially lucrative business opportunities for companies facing the enormous costs of AI development.
Mira Murati, Chief Technology Officer at OpenAI, explained to the Financial Times that the company is being particularly cautious with the launch of the o1 models, as they possess extensive capabilities. However, the models will be made accessible through ChatGPT for paying subscribers and via an API for developers. Murati added that the models have undergone extensive safety testing and performed significantly better in terms of safety compared to previous versions.
