[AI Hanip News] Forget Me, AI... Google Is Running Machine Learning

Editor's NoteSince the emergence of ChatGPT, the world has been buzzing with artificial intelligence (AI). Technology is changing daily, and news is pouring in nonstop. Even top industry experts are struggling to keep up. It's overwhelming to consume everything, but you can't afford to miss out either. We bring you the key points from the weekly AI issues, carefully selected and easy to digest.

Google: "Online is Bard's Playground"

[Image source=Yonhap News]

Google has officially announced that it will use information gathered online to train its ChatGPT rival, 'Bard.' While data has been used for AI training before, the difference now is that the use of data has become more specific?from 'language models' to 'AI models'?and has expanded beyond 'Google Translate' to include 'Bard' and 'Cloud AI.' What draws particular attention is the scope of information Google is collecting. This includes not only information generated when users use Google services (such as emails or documents created, photos uploaded, YouTube comments, etc.) but also user activities (videos watched, purchases made, apps and devices used), user location, movement paths, activities conducted there, and even information about nearby devices or objects?all being intensely studied by AI.

Privacy Invasion? Learning How to Delete Data

Has AI been trained too much? Google is also launching a 'Machine Unlearning Challenge.' While machine learning (ML) is well-known in AI, machine unlearning might be unfamiliar. Machine unlearning involves erasing specific data from the AI's learned dataset without affecting the overall intelligence or learning model. It's similar to forgetting the memory of solving a particular problem but retaining the concept. As AI indiscriminately learns data, issues like privacy invasion and copyright problems have emerged, leading to this new challenge.

ChatGPT's Popularity Declines

Is the era of ChatGPT, the epicenter of the generative AI boom, coming to an end after just six months? The popularity of ChatGPT, which appeared like a comet last November, has waned. Traffic analysis firm SimilarWeb reported that global traffic to ChatGPT dropped 9.7% in June compared to the previous month. Unique visitors decreased by 5.7%, and the time users spent on the website fell by 8.5%. Downloads of Microsoft's Bing app, which incorporates ChatGPT, plunged by 38%. This is the first time various ChatGPT metrics have declined since the service's launch.

OpenAI: "Superintelligence Within 10 Years... Protecting Humanity with 'Avengers'"

OpenAI, the developer of ChatGPT, predicts that 'superintelligence' technology threatening humanity will emerge within 10 years and has decided to gather 'heroes' to control it. OpenAI warns that superintelligence, AI surpassing humans, could potentially cause human extinction. To prevent this, research on AI control technologies, called alignment, is necessary. OpenAI plans to create a 'Super Alignment Team' to protect humanity from superintelligence. Additionally, it intends to allocate 20% of its computing power over the next four years to solving superintelligence-related issues.

European Companies Say "AI Regulations Are Too Harsh"

[Image source=Yonhap News]

European companies have pushed back against AI regulations. More than 150 companies, including Renault, Heineken, and Airbus, sent an open letter to the government. They argued that the EU's AI regulations are too strict and will drive related companies and talent out of Europe. They expressed concerns that this would ultimately undermine technological sovereignty and competitiveness. The regulations the EU is pursuing are indeed stringent. AI models must undergo government risk assessments, and all data used for training must be disclosed.

New York Imposes AI Hiring Regulations: "Must Disclose Gender and Racial Bias"

The United States, which had been relatively lenient on AI technology, has introduced new regulations. New York City has mandated that when AI is used in hiring processes, its biases must be examined. This means checking whether AI favors certain genders or races in hiring. AI hiring is common in U.S. companies, with 83% of them using it. This has led to ongoing fairness controversies. For example, AI trained on male-dominated job sectors might give lower scores to female applicants.

NVIDIA Developing AI Chips for Small Devices

NVIDIA's acquisition of AI startup 'OmniML' was recently reported. OmniML develops software that enables AI to operate on individual devices rather than servers. This allows AI to be used in relatively small devices such as cars, robots, drones, laptops, and smartphones. NVIDIA's existing AI semiconductors have been used to run massive AI models in data centers. For the devices we use daily, only AI-generated results are visible. The day when AI is embedded in every device is not far off.

Self-Evolving Machines Are Coming

AI has designed a central processing unit (CPU). According to a paper published by Chinese researchers, AI designed a CPU with performance similar to the Intel i486SX, a popular CPU from the early 1990s. The design took about five hours, approximately 1,000 times faster than a human. With such technology, self-evolving machines could emerge. Meanwhile, a 'sex toy' incorporating ChatGPT has appeared. Lovense, a sex toy manufacturer, offers products where AI narrates erotic stories based on the user's chosen type.