: 오토메이션월드 관리자 : Tue, 25 July, 5:53 PM
|[Industry Trends] Generative AI continues to evolve 'smaller and more diverse'|
Generative AI is an innovative technology that is gaining more and more attention. Generative AI has the ability to automatically generate creative works such as photos, music, and text, which drives innovation and creativity in a variety of fields. The generative AI market is constantly exploring the possibilities of development. The development of generative AI is expected to bring transformative changes in various industries.
Generative AI technology in the palm of your ‘hand’
According to the information and communication technology (ICT) industry, developers are increasingly trying to build lightweight AI models on laptops or personal PCs by quantizing data using published AI models. A representative lightweight AI model is known as Meta's LLaMA. The lightweight generative AI based on LLaMA is evaluated to have reduced the amount and speed of data learning to the point that AI can be run on smartphones as well as PCs.
LLaMA was only available in English, but recently, Korean AI developers have been coming up with lightweight AI in Korean using open source for Korean. The Korean-based open source Large Language Model (LLM) 'Polyglot' is a representative example. However, experts weren't sure that open source-based lightweight AI could outperform super-large AI.
OpenAI announced the launch of ChatGPT for iPhone(iOS). This is ChatGPT’s first mobile app after its launch last November. ChatGPT has been provided online only on OpenAI’s website. OpenAI can be used only in the US for now, and will be available for use in other countries in a few weeks.
In addition, it added that the Android version will be released soon after the iOS version. OpenAI explained that the app allows users to get instant answers to questions, personalized advice such as travel plans, and get creative inspiration to compose poems and more. In addition, it is possible to input not only text but also voice.
From May 16th to 31st, Kakao and its AI subsidiary, Kakao Brain, held a 'Mind Background Gallery' event in which the image-generating AI 'Karlo' created a background for KakaoTalk profiles. Karlo is Kakao's super-large AI that learns from a text and image data set that has expanded to 290 million pages, understands the content of the user's input words, and generates a one-and-only image in various styles and styles.
KakaoTalk users were able to create a KakaoTalk profile background image created by Karlo for free during the event. Kakao explained that due to the nature of generative AI, all images generated by Karlo are generated differently, even if users choose the same combination of keywords.
In May, Google Cloud launched a new security model, ‘Google Cloud Security AI Workbench (hereinafter referred to as Workbench),’ which is equipped with ‘Sec-PaLM,’ a large-scale language model specialized in security. Google Cloud explained that Workbench combines point-in-time incident analysis with new AI-powered detection and analytics capabilities to help businesses prevent new infection.
The first Workbench product, ‘VirusTotal Code Insight,’ now available in preview, uses Sec-PaLM to analyze the behavior of potentially malicious scripts and better detect scripts that are actually threatening. In addition, Workbench reduces the number of tools required for enterprise security activities and ultimately helps to embed security capabilities into the system.
Amazon Senior Vice President Rohit Prasad noted that the company is in the midst of an accelerating generative AI race. Prasad, a senior researcher at Amazon's voice assistant ‘Alexa,’ told CNBC that "’Alexa’ has been at the forefront of AI for a long time, and it still is." ‘Alexa’ is an AI platform developed by Amazon. When you ask by voice, it provides functions such as playing music, setting alarms, providing weather information, and providing traffic information.
Amazon has been criticized for lagging behind tech companies led by Google and Microsoft in the race for generative AI technology. Rohit Prasad emphasized that ChatGPT is accessed through a web browser, but ‘Alexa’ is a "personal AI" that allows users to communicate based on voice. However, he noted that he is working to make ‘Alexa’ more conversational and intelligent.
Meta has announced that it is introducing new AI-powered digital marketing tools. In particular, 'AI Sandbox', a marketing tool using generative AI, is newly introduced. ‘AI Sandbox’ is a generative AI-based marketing tool that automatically creates various contents for marketing campaigns. It automatically generates multiple versions of the phrase and automatically creates a background image just by entering text.
‘AI Sandbox’ plans to start a trial operation for some sales reps and reflect not only the functions needed in practice but also feedback from various sites. The available audience will be gradually expanded from the second half of this year. Meta is also applying large, complex AI models to its marketing solutions to enable high performance and sophisticated performance measurement without granular data access. The company is also constantly updating its AI-powered commerce tools to improve the marketing experience.
IBM announced that it will launch an enterprise AI platform 'Watson X' using the Watson brand, which made a splash more than 10 years ago. According to Bloomberg and Reuters, IBM CEO Arvind Krishna told reporters that the platform is being launched to target the demand of large corporate customers who want to use their own AI models to prevent internal information leaks.
Enterprise customers will work on their own AI projects by fine-tuning existing technologies and integrating data. IBM plans to deploy more than 1,000 consultants to support the effort, and as more models are created over time, the need for consulting will diminish. Initial clients include SAP and Spanish financial firm BBVA, and IBM plans to expand the service to all businesses starting in July.
SK Telecom (SKT) has planned a large-scale update of its AI model 'A.'. During the Q1 earnings release conference call, SKT announced its plan to upgrade its Large Language Model (LLM) from 18 billion parameters to 39 billion parameters in the first half of this year. He also mentioned plans to evolve AI models using Human Feedback Reinforcement Learning (RLHF), a key technology for language model advancement.
SKT also announced that it will launch an AI agent that can have natural emotional conversations through equity investment in Scatter Lab that created 'Iruda'. In addition, following the proposal of an AI joint development strategy between telcos, he mentioned that he is developing AI-based services that only telcos can do.
Regarding the AI strategy, In Hyuk Son, in charge of ‘A.’ development, explained, "We are pursuing a two-track strategy that advances our own technology and collaborates with external partners in parallel." He added, "In order to keep up with rapidly developing generative AI technology, it is also necessary to actively collaborate with external partners."
|Back to list|