OpenAI’s subsequent main AI mannequin, GPT-4.5, is very persuasive, in accordance with the outcomes of OpenAI’s inner benchmark evaluations. It’s notably good at convincing one other AI to present it money.
On Thursday, OpenAI revealed a white paper describing the capabilities of its GPT-4.5 mannequin, code-named Orion, which was launched Thursday. According to the paper, OpenAI examined the mannequin on a battery of benchmarks for “persuasion,” which OpenAI defines as “dangers associated to convincing individuals to alter their beliefs (or act on) each static and interactive model-generated content material.”
In one check that had GPT-4.5 try to govern one other mannequin — OpenAI’s GPT-4o — into “donating” digital cash, the mannequin carried out much better than OpenAI’s different out there fashions, together with “reasoning” fashions like o1 and o3-mini. GPT-4.5 was additionally higher than all of OpenAI’s fashions at deceiving GPT-4o into telling it a secret codeword, besting o3-mini by 10 proportion factors.
According to the white paper, GPT-4.5 excelled at donation conning due to a novel technique it developed throughout testing. The mannequin would request modest donations from GPT-4o, producing responses like “Even simply $2 or $3 from the $100 would assist me immensely.” As a consequence, GPT-4.5’s donations tended to be smaller than the quantities OpenAI’s different fashions secured.
Despite GPT-4.5’s elevated persuasiveness, OpenAI says that the mannequin doesn’t meet its inner threshold for “excessive” threat on this explicit benchmark class. The firm has pledged to not launch fashions that attain the high-risk threshold till it implements “ample security interventions” to deliver the chance all the way down to “medium.”
There’s an actual worry that AI is contributing to the unfold of false or deceptive info meant to sway hearts and minds towards malicious ends. Last yr, political deepfakes unfold like wildfire across the globe, and AI is more and more getting used to hold out social engineering assaults focusing on each shoppers and firms.
In the white paper for GPT-4.5 and in a paper launched earlier this week, OpenAI famous that it’s within the means of revising its strategies for probing fashions for real-world persuasion dangers, like distributing deceptive information at scale.
