Mistral’s most recent LLM model, which is accessible in an open-source format, comes amid recent releases of models from OpenAI and Google.
Mixtral 8x22B, a new large language model (LLM), and Mistral’s newest effort to take on the big boys in the AI space were unveiled on Tuesday by the French AI startup. Gigazine reports that Mixtral 8x22B is predicted to surpass Mistral’s previous Mixtral 8x7B LLM, which in turn demonstrated signs of surpassing both OpenAI’s GPT-3.5 and Meta’s Llama 2.
A 65,000-token context window, or the amount of text an AI model can analyze and reference at once, is one of the new Mixtral model’s many impressive features. In addition, the parameter size of Mixtral 8x22B, which denotes the quantity of internal variables the model employs in decision-making or prediction, can reach 176 billion.
Mistral is an open-source AI model company founded by researchers from Meta and Google. Here, Mixtral 8x22B can be downloaded as a 281GB file and used by everyone. Simply copy and paste the magnet URL from Mistral AI’s X post into your preferred BitTorrent software to accomplish this yourself.
The AI market is looking for fresh, creative models right now, which coincides with the release of Mistral’s most recent LLM.
The most recent GPT-4 Turbo model with vision capabilities, GPT-4 Turbo with Vision, was made available by OpenAI on Tuesday. It can work with user-uploaded photos, sketches, and other pictures. Google gave developers access to its cutting-edge Gemini Pro 1.5 LLM on the same day, with a free version that allows for up to 50 queries every day. Meta disclosed that its Llama 3 model would make its appearance later this month, not to be outdone.
Frontier models, which can manage a broad range of duties and demands, are what Mixtral 8x22B and these other sophisticated LLMs are known as. With more innovative technology, they hope to surpass earlier versions while evoking the Wild West. There’s a sense of peril associated with the term “frontier.” The dangers of frontier models were discussed by OpenAI in a blog article from July 2023.
“Frontier AI models pose a distinct regulatory challenge: dangerous capabilities can arise unexpectedly; it is difficult to robustly prevent a deployed model from being misused; and, it is difficult to stop a model’s capabilities from proliferating broadly,” stated OpenAI. “One crucial first step is industry self-regulation. To establish norms and guarantee adherence to them, nevertheless, more extensive social discourse and government action would be required.”
The Guardian reports that Mistral’s open-source strategy has drawn some criticism as well. The firm is unable to stop its systems from being misused because anyone may download and improve upon its AI models. Furthermore, if certain biases or defects arise that need to be fixed, the models cannot be taken offline.
