Not so funny? AI safety conference hails new chatbot

In a month when the world’s first AI safety conference took place in Bletchley Park, home to a crack-team of code-breakers during World War Two, and countries across the globe including China and the US signed statements that showed their dedication towards minimising the risks of AI, it was strange that one speaker had anything other than safety on his mind.

Elon Musk was a headline act at the conference, but his focus was apparently less centred around safety and more around fun, as he announced the up-coming release of his new generative AI tool Grok.

Trained on Twitter

Designed to bring a flavour of The Hitchhikers Guide to the Galaxy back to the invention of technologies, this new chatbot won’t be quite the same as ChatGPT.  On its release it will only be available to Premium+ members of X (formerly known as Twitter). Grok is also the only chatbot that is trained on X’s back catalogue, which Musk believes to be a sign that it will only get more entertaining over time. But for anyone who has ever witnessed a Twitter argument online, this approach might already §be setting off alarm bells.

Musk’s approach to X, since his takeover, has been much criticised as he systematically dismantled a significant proportion of the safety features of the site, sacking content moderators under the guise of encouraging ‘free speech’. While many might support his approach, it cannot be denied that this has had real consequences for the platform. As the New York Times reported, since Musk’s takeover of the platform there has been a radical increase in hate-speech on the platform.

So, if this is Musk’s approach to limiting online harms on social media, can we expect his approach to AI safety to be any better? The odds are looking slim. In Musk’s announcement about his new chatbot on X, he appears to show the chatbot offering advice on how to make cocaine. Whilst this is quickly followed up with a line about not taking its advice, there are questions to be asked about the safety of releasing such a tool into the public domain.

‘Not everybody is in the right frame of mind to understand sarcasm…’

AI tools can after all have a significant impact on those using them. For example, the BBC has reported that a man faces nine years in prison for treason after his AI chatbot girlfriend allegedly encouraged him in his attempt to kill the Queen. While an extreme example, this illustrates the ways in which not everybody is in the right frame of mind to understand sarcasm or override a quick ‘don’t do this at home’ message dropped in at the end of some sarcastic advice from an AI chatbot.

How safe Musk’s chatbot will be, and how funny, is still up in the air. Only time will tell what kind of impact its launch will have on the wider world.

Read Echo Callaghan’s latest Technology Roundup in our Innovation Zone here.

To access premium content in the Innovation Zone, join WORKTECH Academy here.