WBA service main screen. Image=Provided by FriendlyAI
AI technology specialist company FriendlyAI (CEO Jeon Byeong-gon) officially launched the K-language model comparison experiment platform, 'WBA (World Best AI, Waba),' on the 6th, allowing even non-experts in AI to participate.
WBA is a service that allows users to evaluate AI language models through a blind test method. The evaluation leaderboard is also disclosed. It is characterized by its easy-to-use interface, fairness, and elements of fun.
Recently, domestic AI companies such as LG AI Research Institute, Upstage, SKT, and Naver have released various language models as open source. However, it is difficult to compare which model has superior performance in real-use environments based solely on some benchmark scores disclosed by the developers.
Therefore, WBA adopts a user-centered evaluation system. The usage is simple. When a user inputs a desired question into WBA, two randomly selected language models immediately provide responses. If the 'logical response' option is checked, two reasoning language models generate responses. The user then selects the response they prefer. The names of the evaluated models are revealed only after the user voting ends, making score manipulation impossible. The WBA leaderboard ranking is determined by summing these voting scores.
WBA model evaluation interface example. Image=Provided by FriendlyAI
The WBA service can be accessed freely by anyone on the homepage without any burden. As responses from two models are generated simultaneously, users can experience various models they have not tried before.
The national AI elite team (independent AI foundation model project) selected by the government on the 4th, including LG AI Research Institute, Upstage, SKT, and Naver, can also be encountered. Additionally, representative models from global big tech companies such as OpenAI, Anthropic, and Google, as well as famous overseas open-source models like DeepSeek and Qwen, are registered, allowing for direct comparative evaluations between the elite team's K-language models and overseas language models.
Jeon Byeong-gon, CEO of FriendlyAI, stated, "As the competition for developing AI language models in Korea has begun in earnest, this is an opportunity to verify what the truly high-performance AI chosen by the public is," adding, "WBA allows anyone to participate and gain fun and fulfillment. Do not hesitate to participate in this K-language model comparison experiment."
ⓒ dongA.com. All rights reserved. Reproduction, redistribution, or use for AI training prohibited.
Popular News
