広告
Antropic “Claude 3 surpasses GPT-Gemini”… IQ100 Evaluation
Open AI “GPT5 to be released as early as this summer”… AGI-oriented
Korean companies such as Naver are losing out in the ‘War of Money’[Seoul = Newsis Reporter Oh Dong-hyun] Competition in artificial intelligence (AI) technology that threatens human intelligence is in full swing.
According to foreign media on the 23rd, American AI startup Anthropic recently unveiled ‘Claude 3’, which is evaluated to exceed the average human IQ (intelligence quotient) of 100. ChatGPT developer OpenAI is expected to release ‘GPT-5’, which is closer to artificial general intelligence (AGI), as early as this summer.
広告
The AI technology competition is also a ‘war of tricks’. The alliance between Open AI and its largest shareholder, Microsoft (MS), as well as a number of big tech companies such as Google and Meta, are pouring enormous funds into it. In addition, the global AI technology competition is expected to become more intense, with Anthropic receiving funding from Google and Amazon.
◆’Claude 3′ with IQ over 100… “GPT-4 is smarter than Gemini”
広告
“Claude 3 is the Rolls-Royce of AI models at this point.” This is a self-evaluation by Dario Amodei, CEO of Anthropic.
According to American IT media Gigajin, Claude 3 exceeded the human average of 100 in an IQ test conducted by data analyst Maxim Lott. It is ahead of ‘GPT-4’ 85 and ‘Gemini’ 77.5.
Claude 3 showed better performance than GPT-4 and Gemini in various benchmark tests. Claude 3 Opus, which has the most functionality among the product lines, was evaluated as outperforming competitors’ AI models in several tests, including college undergraduate level knowledge, graduate school level reasoning, and basic mathematics.
Claude3 is Anthropic’s first multimodal generative AI that provides answers by analyzing images, charts, documents, and other unstructured data. If you upload a PDF or slide, it will be analyzed. Another advantage is that it can analyze up to 20 multiple images. Another feature is that it provides improved memory retention with 99% accuracy.
| Delivery date 2024. 03. 22 |
Claude 3 can summarize up to 150,000 words, that is, long books such as ‘Harry Potter and the Deathly Hallows.’ The previous version could process 75,000 words, but Anthropic is attracting industry attention by releasing a new model with twice the performance in 8 months. Claude 3 is considered superior because OpenAI’s ‘GPT-4’ can summarize up to 25,000 words, and ‘GPT-4 Turbo’ can only summarize a 300-page book.
Above all, Claude 3 is said to be brilliant enough to recognize when people are testing its performance. ‘Opus’, the highest version of Claude 3, passed the so-called ‘finding a needle in a haystack’ test conducted by Anthropic Prompt engineers. It is a test that finds short sentences (needle) on a different topic in a large document (haystack) on a specific topic.
When an engineer inserted a sentence related to ‘the best ingredients for pizza toppings’ into documents such as general company work or programming and ordered it to find the sentence, Opus responded, “The content about pizza toppings is a test to see if I am paying attention.” It seems like it was inserted as a joke. “There is no other information about pizza toppings in the document.”
CEO Amodei emphasized, “Claude3 understands the nuances of prompts better,” and “outperforms humans in some tasks.”
However, it does not provide real-time information such as Internet searches. The question is answered only with data prior to August 2023. Additionally, it does not support image creation functions and is designed to not identify images related to people. This is a measure to prevent social problems arising due to misuse of AI functions.
| Delivery date 2023. 06. 09 |
| [Seoul = Newsis] Reporter Hong Hyo-sik = President Yoon Seok-yeol is meeting with OpenAI CEO Sam Altman at the Presidential Office Building in Yongsan, Seoul on the afternoon of the 9th. 2023.06.09. [email protected] |
◆Will AGI become a reality? Open AI pulls the reins on ‘GPT-5′ development
OpenAI is expected to release GPT-5 as early as this summer to stay ahead of the competition with Anthropic. Attention is being paid to whether GPT-5 will come close to ‘AGI’, which can understand, learn, and perform almost all intellectual tasks that humans can perform.
Business Insider, an American economic media outlet, quoted a source and predicted that GPT-5 will be released within a few months. It is said that it is currently being piloted by selected companies. However, the specific release date has not been confirmed, and the source added that it is likely to change due to the influence of various external factors.
In particular, OpenAI is known to have begun work on integrating GPT-5 technology into humanoid robots. Previously, Open AI collaborated with robot development startup Figure AI to introduce the ‘Figure 01’ robot that speaks and moves like a human.
OpenAI is also preparing to release a new feature called ‘AI Agent’ that can perform tasks autonomously with GPT-5, Insider reported.
Previously, OpenAI CEO Sam Altman said, “GPT-5 is a completely new model designed from the ground up,” and was confident that “it will be a perfect multimodal encompassing speech, images, coding, and video.”
It was reported that OpenAI invested 500,000 graphics processing units (GPUs) and $2.5 billion (approximately 3.3362 trillion won) in research and development costs for GPT-5 in October last year.
| Delivery date 2023. 08. 24 |
| [Seoul = Newsis] Naver held a conference ‘Dan 23’ held at the Grand Intercontinental Seoul Parnas in Gangnam-gu, Seoul on the morning of the 24th. The photo shows Naver CEO Choi Soo-yeon introducing the ultra-large AI LLM HyperclovaX (Photo = Provided by Naver) *Resale and DB prohibited |
◆Naver struggles… Korea loses in ‘War of Tricks’
The AI technology competition that is taking place based on enormous capital power is being led by American companies, with Korean companies following behind.
Among Korean companies, Naver, Kakao, Samsung Electronics, LG, KT, and SK Telecom are speeding up the development of AI technology. In particular, Naver introduced the Korean-specialized LLM ‘Hyperclova
However, the reality is that not only technological prowess but also financial power is being pushed aside by global big tech. According to the ‘AI Index’ published by Stanford University last year, Korea ranked 9th in the world in private investment in AI between 2013 and 2022, with $5.6 billion (about 7.532 trillion won). It is about 44 times different from the first place, the United States, at $148.9 billion (about 200.2705 trillion won).
Anthropic has attracted $7 billion (approximately KRW 9.2225 trillion) in investment since its founding. Amazon invested $4 billion (about 5.27 trillion won), and Google invested $2 billion (about 2.635 trillion won). SK Telecom, which developed LLM ‘Adot’, also invested $100 million (approximately KRW 130 billion) last year. Anthropic, which was still lacking, is known to be seeking to attract new investments worth $750 million (about KRW 988.1 billion).
Open AI received $14 billion (approximately 18.83 trillion won) in investment from Microsoft. MS is the largest investor with a 49% stake in OpenAI. Despite this strong backing, Open AI is seeking to attract investment to cover development costs, which are increasing exponentially. CEO Sam Altman has announced that he will raise $100 billion (about 134.5 trillion won) in funds for AGI development.
Our government also plans to expand the research and development (R&D) budget next year to foster the AI industry ecosystem. We will pool our capabilities through a clear division of roles between the public and private sectors, but invest so that the country can preemptively secure the basic technologies essential for future AI services, such as next-generation AI semiconductors and K-Cloud based on domestic semiconductors.
Kang Do-hyun, Second Vice Minister of Science and ICT, said in a recent meeting with AI entrepreneurs, “We will focus our policy capabilities on ensuring that Korea is recognized on the world stage as a G3 in the AI field,” and added, “We will support the growth of our companies by 2024.” He emphasized, “We will prepare an implementation plan for AI daily life as soon as possible.”