Chinese AI Tops Hugging Face's Revamped Chatbot Leaderboard 9
Alibaba's Qwen models dominated Hugging Face's latest LLM leaderboard, securing three top-ten spots. The new benchmark, launched Thursday, tests open-source models on tougher criteria including long-context reasoning and complex math. Meta's Llama3-70B also ranked highly, but several Chinese models outperformed Western counterparts. (Closed-source AIs like ChatGPT were excluded.) The leaderboard replaces an earlier version deemed too easy to game.
This isn't really open source (Score:1)
Headlines getting silly... (Score:3)
Re: (Score:2)
I bet it was created by AI. ;)
Re: (Score:2)
HuggingFace is like Github for AI. It's *massive*. If you do anything with AI, you know HuggingFace.
I assume that's the only part you had trouble with?
Re: (Score:2)
Hugging Face offers AI hosting as a service.
One interesting aspect that isn't really covered in TFA is that there are so many open source Chinese AIs now. Open source is being embraced very quickly there, spurred on by a court decision last year that affirmed that the GPL is valid and enforceable. It's created an environment where small players can compete without needing massive engineering teams.
The leaderboard makes no sense (Score:3)
Could never make heads or tails of it. My favorite models are sometimes not even listed and not so great models routinely show up at the top. What is phi-3 doing at the top? Where is DeepSeek v2?
Still amazed to see how much of a difference tuning makes.
Re: (Score:2)
I think they only include models that Hugging Face offers as a service. Their business is hosting open source AI models for you, so you don't have to buy an expensive GPU or ten.