By the time a taxi driver starts talking about investing in a certain stock, it’s usually too late. But when a taxi driver tells me he no longer trusts what he reads online, I think to myself: it’s about time.
Everything is distorted online. Journalists cherry-pick quotes, facts, events, and trends, while netizens become masters at crafting alter egos – so much so that you might not recognize them if you met them in the street.
AI makes this problem even worse. The internet, which once at least resembled real life, is becoming more like science fiction, where everything – from a lawsuit to a human being – might be nothing more than a hallucination.
OpenAI has just updated its Preparedness Framework, which outlines how the company monitors and prepares “for advanced AI capabilities that could introduce new risks of severe harm.”
Essentially, OpenAI is trying to ensure that its large language models aren’t used to create biological weapons or assist malicious hackers, among other threats.
What Fortune noticed is that mass manipulation and disinformation are no longer classified as critical risks. This means OpenAI will stop assessing its AI models for risks that they could persuade or manipulate people. Instead, those risks will now be addressed through terms of service, restricting AI’s use in lobbying and political campaigns.
Since large language models can be jailbroken and exploited for malicious purposes like writing malware, no one really expects an in-house assessment before a model’s release to be a silver bullet.
However, it seems we’re slowly getting used to a new reality where copyright lines blur, conspiracy theories are amplified by AI and automation until many start to believe them, and taxi drivers no longer trust what they read or hear.
Perhaps they shouldn’t – but this growing sense that everything around us is fake is pushing people toward conspiracies, and that’s where the real problems begin.
Your email address will not be published. Required fields are markedmarked