Ask AI about Lee Kang-in VS Son Heung-min virtual match

juimunijui9

Hot soccer players these days are Lee Kang-in and Son Heung-min. Although it is difficult to achieve in reality, we asked Korea's hottest conversational generation AI about the results of a virtual match. Criteria for selecting AI that will predict the winner We first considered conversational AIs and asked LLM (Large Languge Model)-based chatbot services for the results. Among these, we selected the five AIs that are being discussed most in Korea. Specifically, they are Google's Bard, Microsoft's New Bing AI, Meta's Llama2, Chat GPT 3.5, and Claude.ai. If possible, we tried to proceed with a version with the latest data as of July 2023. Introducing conversational generative AI that will judge virtual matchups Source: Bing – Image Creator Google Bard It is a lightweight version of the service based on the Google search engine's LaMDA (Language Dialogue Model) and PaLM (Natural Language Processing Model). The advantage is quick response speed and up-to-date information. There is a story that after the humiliation at the demonstration, originality was greatly limited to ensure accuracy of answers. Korean was added in May 2023, and will be available in 46 languages ​​from July 2023. You can also receive answers in image form from May 2023.

Microsoft New Bing AI It is said that the Prometheus model based on GPT-4 was used. Q&A is conducted using the chat function and is only available in the Microsoft Edge browser. (Of course, it cannot be used on competitor Google Chrome.) It has the advantage of applying the  Special Data  latest data, but currently, the search results are displayed interactively, so the answer seems to be mechanical. Meta – Llama 2 (served by Perplexity Labs) Meta is open source and is being distributed free of charge for both research and commercial use. In 2023, Llama 2 (Llama 2) open source was released. Three versions of the model were released, pre-trained and fine-tuned with 7B (7 billion), 13B (13 billion), and 70B (70 billion) parameters.




In this match, we used the Llama-2-13B version provided by Perplexity Labs. Chat GPT 3.5 It is synonymous with interactive generative AI. This time I used version 3.5, which is free. The biggest weakness is that the data standard is before September 2021, so the latest data is not reflected. However, there are many evaluations that a dramatic improvement has been made in the GPT 4 version. So, I think this match should be considered as an evaluation from the intermediate version. Cl aude.ai This is a service from Anthropic, an American startup created by former members of OpenAI, the developer of Chat GPT. When OpenAI announced that it would charge a fee, it was established as a public interest company in response to this. Currently only available in select countries, including the US and UK. Source: Bing – Image Creator Criteria for judging player abilities Since it is not possible to check all detailed abilities, we decided to use abilities that can be summarized in a broad range as a standard.