Upstage "Self-developed AI Surpasses ChatGPT Performance"

Surpassing GPT-3.5 in Hugging Face Evaluation Scores

Upstage announced on the 1st that its generative artificial intelligence (AI) model scored 72.3 points on the 'Open Large Language Model (LLM) Leaderboard' operated by Hugging Face, the world's largest machine learning platform, surpassing the performance of GPT-3.5, which is the basis of ChatGPT, and taking first place.

The Hugging Face Open LLM Leaderboard is considered a benchmark for open-source generative AI models. Over 500 open models worldwide compete based on the average scores of four indicators: inference and common sense abilities, comprehensive language understanding, and hallucination prevention.

[Photo by Upstage]

Last month, the 3-billion-parameter model released by Upstage through Hugging Face scored an average of 67 points. On the same day, it surpassed Meta's 'LLaMA 2' 70-billion-parameter model and achieved the first place for the first domestic LLM. Following this, Upstage released a model fine-tuned with more data on the latest LLaMA 2. As a result, the Upstage 70B model recorded a score of 72.3 on the leaderboard evaluation. This exceeded the benchmark score of GPT-3.5 (71.9 points), which is the foundation of ChatGPT.

Seonghoon Kim, CEO of Upstage, said, "We are pleased to confirm that Upstage's generative AI model demonstrates world-class technological capabilities by outperforming ChatGPT." He added, "Going forward, Upstage will accelerate strengthening its dominance in the domestic and international private AI markets based on overwhelming technological prowess."

This content was produced with the assistance of AI translation services.