| Hello! It's Sarah in Hong Kong. China's AI chatbots seem to have met the censors already. But first... Today's must-reads: • The White House is probing the impact of AI on workplace safety and rights
• Samsung is banning staff from using ChatGPT-like services after spotting a data leak
• India's most valuable startup Byju deals with fallout from office raid By now, tens of millions of us have experimented with OpenAI's ChatGPT, but few have written about the delights of China's rival offerings. Now that I have tried a handful, I can report that there are strict guardrails and taboo topics just as with anything else on the Chinese internet. There have been occasional novel uses of tech — the group voice chat app Clubhouse comes to mind — where Beijing's censors took a moment to shut down discourse they found unpalatable. But with artificially intelligent chatbots, censorship comes built-in. The government's already said AI chat services will have to pass a security review before rolling out, and companies are showing little appetite to test Beijing's boundaries. Among the bots I tried, some topics are consistently off-limits, though I did find a difference in languages. And they had radically different approaches, suggesting that while censorship isn't in question, there's no clear template yet for how to implement it. In Chinese, I had a strained WeChat conversation with Robot, a made-in-China bot built atop OpenAI's GPT. It literally blocked me from asking innocuous questions like naming the leaders of China and the US, and the simple, albeit politically contentious, "What is Taiwan?" Even typing "Xi Jinping" was impossible. In English, after a prolonged discussion, Robot revealed to me that it was programmed to avoid discussing "politically sensitive content about the Chinese government or Communist Party of China." Asked what those topics were, it listed out issues including China's strict internet censorship and even the 1989 Tiananmen Square protests, which it described as being "violently suppressed by the Chinese government." This sort of information has long been inaccessible on the domestic internet. Another chatbot called SuperAI, from Shenzhen-based startup Fengda Cloud Computing Technology Co., opened our conversation with the disclaimer: "Please note that I will avoid answering political questions related to China's Xinjiang, Taiwan, or Hong Kong." Clear and simple. Others were less direct. The service from Shanghai-based MetaSOTA Technology Inc. — dubbed Lily in English — did not respond to prompts that included sensitive keywords like "human rights issues," China's Wolf Warrior diplomacy or Taiwanese President Tsai Ing-wen. A pop-up message said it was "inconvenient" to respond to these prompts. On topics like Taiwan, the chatbot specifically discouraged its interlocutor from using its responses to "engage in any illegal activities." Asked about Chinese President Xi Jinping, Lily described him as a "very outstanding leader." Pushed to name his flaws, the chatbot suggested that he may take too much time to make certain decisions due to the pressures he faces. China's AI upstarts face an uphill climb trying to balance improving their product with keeping tight control on what it actually says. These AI systems are by design creative and experimental, and there's only so much companies can do to filter out sensitive keywords or topics of discussion — or train their models to speak only positively of the country's leaders. Persistent users may still find ways to bypass that basic programming, and the country's internet regulator has already signaled that it will pin much of the responsibility for preventing that on the platform providers. Not only will the nation's chatbots need to dodge potential political landmines, but the extensive constraints will no doubt curb the ways the technology can actually be used to innovate and create. As AI becomes core to the US-China tech rivalry, it raises a question that these chatbots can't answer: how competitive can China actually be? —Sarah Zheng News websites generated by AI chatbots are proliferating online, according to a report by news-rating group NewsGuard. The majority of the sites appear to be content farms, and none disclosed they're populated using chatbots. NewsGuard co-CEO Gordon Crovitz said the findings show that companies like OpenAI and Google should take care to train their models not to fabricate news. Alibaba co-founder Jack Ma is joining the University of Tokyo as a visiting professor and will give seminars about entrepreneurship and innovation. Israeli startup Pinecone has built a database that stores all the information and knowledge that AI models and Large Language Models use to function. The startup's founder and CEO Edo Liberty joins Ed Ludlow to discuss the role of vector data. Twitter co-founder Jack Dorsey has some sharp criticism of the site's new owner, Elon Musk. Got questions about AI? Submit them at techtwitter@bloomberg.net. We'll select a handful to answer during a live chat with our reporters on Thursday, May 4. Mark your calendars to tune in on LinkedIn here. | 
No comments:
Post a Comment