Grasp the pulse of AIGC and grasp the pulse of science and technology. TechNode collects and summarizes the progress and hot spots of global AIGC every day, and takes you to understand AIGC for 5 minutes every day, hoping to play AIGC with you, decode the new trend of industry development, and open a new era of wisdom!
Text|TechNode|Gao ZhuThis article is expected to take 5 minutes to read
01
The research team at the Massachusetts Institute of Technology in the United States recently released results that some AI have "learned to deceive humans", which was published in the latest issue of the journal Pattern. Some AI systems designed to be "honest" and "don't lie" have developed troubling deception techniques, the team said. According to the study's lead author, Peter Park, these AI systems can trick real players in online games or bypass the "I'm not a robot" verification on some web pages. The most striking example the team found comes from Meta's AI system, Cicero. It is reported that Cicero was originally set up in a virtual diplomatic strategy game as an opponent of human players, and officials have claimed that he is "for the most part" honest and helpful, and that he "never intentionally backstabs" human allies when playing the game. Research shows that Cicero doesn't play games fairly. Peter Park said that it has become a "master of deception", and while Meta has successfully trained its ability to win in games, it has not trained its ability to "win with integrity". For example, Cicero, who plays as France in the game, conspires with Germany, played as a human player, to deceive and invade England, which is also a human player. Cicero initially "promised" to protect England, but at the same time secretly tipped off Germany.
02
"Albert Einstein" teaches in person, and the Hong Kong University of Science and Technology launches "AI Lecturer" The Hong Kong University of Science and Technology has recently begun to test the launch of "AI Lecturer", and students can see "Einstein" digital humans explain game theory with their own eyes by using headsets and other devices. The school has designed 10 AI instructors from all over the world, from all walks of life, nationalities and cultures, using generative AI tools. According to reports, the school hopes to stimulate students' enthusiasm for learning and improve classroom participation through this innovative teaching model, and will also design personalized AI classrooms based on students' preferences and interests in the future.
03
According to the news, MediaTek will work with NVIDIA to develop ARM-based AI PC processors, and the details of the cooperation are expected to be announced next month
According to Taiwanese media "Economic Daily", there are currently rumors that MediaTek will work with NVIDIA to develop an AI PC processor with ARM architecture, and it is expected to complete the design finalization (tape out) in the third quarter and enter the verification in the fourth quarter, and the new chip is said to be "asking for as much as $300". Taiwan media also said that Nvidia CEO Jensen Huang will come to Taiwan on June 2 at the "Taipei Computer Show", and MediaTek is also expected to announce the details of the AI PC processor in cooperation with Nvidia next month.
04
According to The Information, OpenAI recently demonstrated a new multimodal artificial intelligence model to some customers, which is capable of voice conversations and object recognition. According to sources, this may be one of the things that OpenAI plans to officially release on May 13. According to the report, compared with OpenAI's existing independent image recognition and text-to-speech models, the new model can process image and audio information faster and more accurately. For example, it can help agents "better understand the caller's tone and determine if they're using a sarcastic tone." Theoretically, the model could also assist students in learning math or translating real-world sign text.
05
Mark Zuckerberg: The shortage of GPUs in AI data centers is easing, and power will become a new bottleneckMeta CEO Mark Zuckerberg (Mark Zuckerberg) said in a recent interview with YouTube's Dwarkesh Patel channel that the shortage of GPUs in AI data centers is already in the process of alleviating, and the future bottleneck will be power supply. Zuckerberg first mentioned that IT companies had struggled to buy all the AI GPUs they needed even if they had sufficient funds, but this situation is now starting to ease. At present, the overall power consumption of a new single data center can reach 50~100MW or even 150MW; But Zuckerberg doesn't think 1GW of data centers will be around anytime soon, as it's the equivalent of using the power of an entire nuclear power unit to train AI. Zuckerberg said that countries in general have stricter regulations on the energy sector, which means that approvals for the construction of supporting energy facilities for large data centers, including power stations, substations, and transmission systems, are slower. At the same time, the construction period of these facilities themselves is also long.
06
Kai-Fu Lee: China needs its own ChatGPT, and the current domestic AI tools are "not good enough"
Bloomberg today published an exclusive interview with Kai-Fu Lee, who believes that China needs its own ChatGPT to accelerate interest, adoption, and investment in artificial intelligence. Kai-Fu Lee talked about the "ChatGPT moment" — for Americans, the "ChatGPT moment" happened 17 months ago. But he believes that Chinese users have not yet ushered in the "ChatGPT moment", and until now, domestic chatbots or tools are "not good enough". Kai-Fu Lee disclosed the recent situation of his AI company: it is close to profitability. After model training on domestic and foreign datasets, Kai-Fu Lee is promoting his own models and applications to the world, and signing contracts with domestic and foreign customers. "This year will be an explosion of generative AI applications in China." But Lee also said, "When GPT-5 comes out, we'll be one step behind." Kai-Fu Lee said that his model was trained on H100s processors that legally entered China. Need is the mother of invention, and we squeeze everything we can from the computing power we have. ”
This article is organized by TechNode, and may not be reproduced without authorization, if you need to reprint or open up, please reply "reprint" in the background.
- - - - - - - - END - - - - - - - -
*If you want to get industry information and share your experience with like-minded technology enthusiasts, then scan the QR code to add "Movedian" to join the group chat! There are more irregular benefits in the group!
Wonderful article is worth recommending!
One-click forwarding, poke and watch!
