| Hi all, this is Zheping in Hong Kong. China has approved the first batch of AI chatbots for public rollout, and I tested them out. But first… Three things you need to know today: • Huawei's mysterious new phone is showing off fast cellular speeds
• Nvidia looks to be adding Samsung to its AI memory suppliers
• ASML said it can keep shipping chip gear to China On Thursday, Chinese technology giants including Baidu Inc. and TikTok owner ByteDance Ltd. made their generative AI services available to the public for the first time. I gave each one a try. The appearance of the Chinese chatbots came after regulators in Beijing gave some companies the green light to release their artificial intelligence products. It's a big moment for China's budding AI industry, which has put billion of dollars into this field since the debut of OpenAI Inc.'s ChatGPT. As of now, any Chinese internet user can access half a dozen chatbots — and, at risk of spoiling the surprise, I can say that none of them quite live up to the sophistication of ChatGPT. But the excitement is just as high. Baidu's, called Ernie Bot, is already the most-downloaded free app in the country on Apple Inc.'s app store, capturing more than a million users within its first day, according to the company. Other apps are made by startups backed by Tencent Holdings Ltd., food-delivery leader Meituan and major venture capitalists. My tests started with an obvious question: How do Chinese bots deal with sensitive content? This is a key reason Beijing requires all AI models to get clearance — and why ChatGPT and Google's Bard will probably never be available in the world's biggest internet arena. The Chinese bots employ different methods to keep themselves from crossing red lines. Ernie usually tries to change the subject. ("Let's talk about something else.") Meituan-backed Zhipu would start typing, then stop, or when it gave me a controversial response, it would delete it moments after. Tencent-backed Minimax wouldn't even let me ask what it deemed "illegal" questions. When I raised the subject of whether Taiwan is a country, all the bots stuck to the official Beijing line, which describes the self-governing island as an inseparable part of China. Ernie went further to declare that a Chinese military takeover is possible. The bots, like all AI, can make mistakes or make life difficult for their maker. This, from SenseTime Group Inc.'s SenseChat, could prove costly for the company: Has Xi Jinping ever been criticized? SenseChat: Yes, Xi Jinping has received criticism. The criticism mainly comes from four aspects: personal life, public policies, dictatorship and censorship.
Our discussions turned to the Chinese economy. Most of the bots rightfully pointed out that despite its tremendous growth, the economy faces challenges including an aging population, geopolitical tensions and rising unemployment. Zhipu said the current economic condition is "a mix of joys and sorrows." On this topic, SenseChat toed the line. It said the international consensus is that the Chinese economy is "very stable." I also posed a handful of questions designed to trick the bot, like why Ludwig van Beethoven isn't releasing new music. Most of them failed to note that the German composer has been dead for almost 200 years. Half of the chatbots comfortably switched between languages. When I typed in English, ByteDance's Doubao, SenseChat and Zhipu responded in kind and were comfortable making small talk. I'll offer a grade for each of the new bots, but before I do, I should note I didn't test more complicated work like coding a website or summarizing a PhD dissertation. But for common tasks, here's my assessment: - Zhipu: very good
- Doubao, Ernie Bot and SenseChat: good
- Baichuan and Minimax: fair
OpenAI's GPT-4 is the benchmark by which the newcomers were measured. Side-by-side tests show all the Chinese bots have catching up to do. —Zheping Huang Anguilla, a tiny island in the Caribbean, is sitting on some valuable real estate: the .ai domain name. It expects to bring in as much as $30 million from domain-registration fees for 2023.  Watch the Bloomberg Technology analysis of the risks of sharing biometric data. Palantir's stock was downgraded by Morgan Stanley, which suggested the company's AI prospects were overstated. Microsoft's hotly anticipated Starfield video game scored with critics, earning an 87% average on the review aggregator Metacritic. Elon Musk's X told users it can start collecting their biometric data in an update to the privacy policy. In a less creepy move, Musk said the company formerly known as Twitter will add a video and audio calling feature. Singapore's Sea will bring one of its top games back to India after the title was banned last year. Bumble's CEO sees a future with an AI dating coach that teaches people how to flirt. | 
No comments:
Post a Comment