The leaked model, labeled “miqu-1-70b,” was shared on HuggingFace and has been praised for its impressive capabilities. Its performance on LLM benchmarks, notably the EQ-Bench, was reported to rival that of OpenAI’s GPT-4. Machine learning researchers have lauded it as one of the best open-source AI models currently available. Maxime Labonne, an ML scientist at JP Morgan & Chase, suggested that fine-tuned versions of the model might soon outperform GPT-4.
Key takeaways:
- Paris-based AI startup Mistral is working on a new version of its open-source AI models that may rival or exceed OpenAI’s flagship GPT 4.
- An older model was leaked on HuggingFace by an over-enthusiastic employee, confirmed by Mistral’s co-founder and CEO Arthur Mensch.
- The leaked model, labeled “miqu-1-70b,” has shown impressive capabilities, with its performance on LLM benchmarks reported to rival that of OpenAI’s GPT-4.
- Machine learning researchers and professionals, including those from JP Morgan & Chase, have praised the model, suggesting it could soon outperform GPT-4.