[AI One-Bite News] OpenAI vs Google: London Battle Foretold

"Surpassed ChatGPT," says China's Baidu
US bans export of low-performance semiconductors

Editor's NoteSince the emergence of ChatGPT, the world has been buzzing with artificial intelligence (AI). Technology is evolving daily, and news is pouring in nonstop. Even top industry experts are struggling to keep up. It's overwhelming to consume everything, but starving is not an option either. We bring you the core highlights from the weekly AI issues, carefully selected and easy to digest.

OpenAI Penetrates Google's Stronghold... Google Counters with Gemini

OpenAI, the American AI startup behind ChatGPT, is entering enemy territory. They have chosen London, UK, as their first overseas office. London is also home to DeepMind, the hub leading Google's AI strategy.

[Image source=Yonhap News]

OpenAI, headquartered in San Francisco, USA, is opening its first overseas office. Why London? For starters, London is a 'hotspot' for the AI boom. According to IT media TechCrunch, as of 2021, there are over 1,300 AI companies in London. It is the city with the largest pool of funding and talent in the UK. Politically, it is also significant. While the European Union (EU) has actively pushed AI regulations, the UK is seeking a third path by prioritizing its own AI development. Google's DeepMind headquarters is also located in London. Demis Hassabis, DeepMind's CEO, is a London native, but he likely considered similar reasons as OpenAI. In April, Google's parent company Alphabet merged its AI research organizations Brain and DeepMind, placing Hassabis in charge. This can be considered the true headquarters directing Google AI.

DeepMind is preparing a counterattack. The massive AI being developed to rival ChatGPT is called 'Gemini.' Gemini recently attracted attention when news broke that it is being trained with YouTube videos. Unlike ChatGPT, which communicates only through text, Gemini can understand images and videos as well. The ability to process different types of information, such as text and images, is called 'multimodal.' Being multimodal doubles its capabilities. For example, it can watch a video and provide a one-line summary or analyze sales ledgers and present them as graphs. CEO Hassabis expressed in an interview, "If everything goes according to plan, Gemini will not only analyze text but also have planning and problem-solving abilities," raising expectations.

Voluntary or Forced?… China's Baidu "Surpasses ChatGPT, Slows Down Development"

China's largest search company Baidu is brimming with confidence. They claim their self-developed AI chatbot 'Ernie Bot' surpasses ChatGPT. Citing experiments reported by the state-run China Science Daily, they stated that Ernie Bot (Ernie 3.5) outperformed ChatGPT (GPT 3.5) in comprehensive ability evaluations. In terms of Chinese language proficiency, it even surpassed ChatGPT's latest version, GPT-4. They said that Ernie Bot evolved at an incredible speed just three months after releasing the previous version (Ernie 3.0), to the extent that future upgrade speed and costs could be reduced.

China may have to slow down AI development, possibly involuntarily. The U.S. government has blocked exports of even low-performance AI semiconductors to China. The Wall Street Journal (WSJ) reported that the U.S. Department of Commerce may tighten semiconductor export regulations to China as early as next month. Chinese companies mainly use NVIDIA's graphics processing units (GPUs). GPUs act as the brain chips in generative AI like ChatGPT. Since last October, the U.S. has not sold GPUs such as the A100 and H100, used by Microsoft (MS), Google, and others, to China. NVIDIA circumvented the regulations by producing lower-performance versions, the A800 and H800, for export to China. Now, the U.S. plans to block these as well. This is a difficult situation for China, which had previously avoided U.S. pressure by using multiple low-performance GPUs instead of one high-performance GPU.

ChatGPT-Written Tweets Outperform Human Ones... U.S. Politicians Ban Its Use

A curious paper was published in the international journal Science Advances. It revealed that people trust tweets written by ChatGPT more than those written by humans. Initially, people could not distinguish whether a tweet was written by a human or ChatGPT. Regardless of the accuracy of the information, ChatGPT was trusted more. This means ChatGPT conveys messages naturally and credibly. Conversely, it shows how powerful a tool AI is for deceiving people. This echoes former Google CEO Eric Schmidt's warning that fake news could wreak havoc on next year's U.S. presidential election.

Beyond fake news, a generative AI alert is spreading in political circles. The U.S. House of Representatives has issued a ban on ChatGPT use in offices starting this week (26th). Free services where input data is used for ChatGPT training are prohibited; only paid versions that do not retain information can be used. This is to prevent information leaks. Unreleased information must not be used even in paid versions, and other chatbots besides ChatGPT are completely banned.