-
OpenAI’s GPT-4o model has been assessed as “medium” risk in a new safety evaluation
-
The assessment was based on four key categories: cybersecurity, biological threats, persuasion, and model autonomy
-
OpenAI employed external experts to identify potential risks and vulnerabilities in the model
OpenAI has released a detailed safety assessment of its GPT-4o model, categorizing it as “medium” risk. The assessment was conducted by internal and external experts, who evaluated the model’s potential risks and vulnerabilities in four key categories. While the model was deemed low risk in most areas, it was found to have a higher persuasive potential than human-written text, leading to the overall “medium” risk rating.
The safety assessment is part of OpenAI’s efforts to be transparent about its AI systems and mitigate potential risks. The company has faced criticism and calls for greater accountability and transparency in its safety review processes. The release of the GPT-4o model, with its advanced capabilities, has raised concerns about potential misuse and misinformation.
OpenAI’s safety assessment is a step towards addressing these concerns and demonstrating the company’s commitment to responsible AI development. However, some experts and lawmakers continue to call for greater transparency and external oversight of AI systems.
