Mistral CEO confirms new open source AI model, comparable to GPT-4, was leaked

The open-source AI community recently experienced a surge of excitement following the release of a new large language model (LLM) named “miqu-1-70b” by a user called “Miqu Dev” on HuggingFace, a leading AI model and code-sharing platform. This model, which matches the interaction format of Mistral’s Mixtral 8x7b and has been compared to OpenAI’s GPT-4, quickly gained attention for its high performance on LLM tasks. Amid speculation over its origins, it was revealed that an early access customer of Mistral inadvertently leaked this version of the model, which was based on retraining from Meta’s Llama 2. Mistral’s co-founder later confirmed this, hinting that an even more advanced version of the model, potentially surpassing GPT-4’s capabilities, might soon be released. This development signifies a potential shift in the competitive landscape of generative AI and poses a challenge to OpenAI’s dominance, embodying a significant moment for the open-source AI community and showcasing the rapid advancements in the field.