What Did OpenAI Do This Week? - 30/07/2023 [+25 LINKS]
OPENAI DITCHES OWN AI DETECTOR…WHAT NOW?
OpenAI has quietly discontinued its AI-writing detector citing a "low rate of accuracy" and provided nothing to replace the tool. The AI classifier launched on January 31st and shut down on July 20th. Designed to help users differentiate between human-written and AI-generated text, it was released amidst an outcry from educators about students potentially using ChatGPT to write essays and schoolwork. Online searches for ‘AI essay writing’ may have skyrocketed to 2,041% and headlines proclaimed doom, “Will everyone become a cheat?”, “The college essay is dead?” but OpenAI never claimed its classifier was fit for purpose. The company readily admitted that it was not "fully reliable," given it correctly identified only 26 per cent of AI-written text as "likely AI-written" and incorrectly labelled human-written works 9 per cent of the time. You might ask why did they ship it at all?
Without a panacea, educators quickly moved to restructure their lessons and assignments, but content detection was and is still a major concern. Any such tool, if truly reliable, will be invaluable given the impacts shown and feared over the last six months from gen AI. OpenAI recognise this and on closure, they made a promise to develop more effective methods for identifying AI-generated audio and visual content.
‘We are working to incorporate feedback and are currently researching more effective provenance techniques for text and have made a commitment to develop and deploy mechanisms that enable users to understand if audio or visual content is AI-generated.’ - OpenAI
Still, doubt is circling about whether any entity can crack an AI identifier. "If OpenAI can't get its AI detection tool to work, nobody else can either. I've said before that AI detection tools are snake oil sold to people and this is just further proof that they are. Don't trust them. They're nonsense." said AI writer and futurist Daniel Jeffries.
Quite a few think they can. An industry of commercial AI detectors has sprouted up over the past six months, and researchers at the University of Kansas now claim to have a method with over 99% accuracy to identify AI-generated scientific writing. Meanwhile, a recent study showed that tools purporting to detect AI-generated text can be easily fooled.
Interestingly though, this was also the week that OpenAI (alongside Microsoft, Google and Anthropic) formed a new industry body, and made a “voluntary commitment” to the White House to develop robust watermarking and/or detection methods. (No coincidence there then?!) But there’s hope that OpenAI given its “considered the best at this AI stuff” has a new plan to answer one of the most important questions the technology industry must answer ‘was this created by AI?’
Want to know what this means for you and +25 news links this week? Subscribe now… Need more reasons? Forbes highly recommends us, we’re listed in the top 100 AI Substacks, and you get a stack of resources and other elements all for just £15 a month or $99 a year. What’s not to like?
Keep reading with a 7-day free trial
Subscribe to What Did OpenAI Do This Week? to keep reading this post and get 7 days of free access to the full post archives.