The past week in the world of artificial intelligence has been abuzz with the release of three major large language models (LLMs). Google introduced Gemini Pro 1.5, OpenAI unveiled a revamped version of GPT-4 Turbo, and Mistral launched the openly licensed LLM Mixtral 8x22B, all within a span of 24 hours. These releases marked significant advancements in AI technology and capabilities, with each model offering unique features and functionalities to users.
Gemini Pro 1.5 by Google, now available in over 180 countries, boasts a powerful public LLM with a free tier allowing up to 50 requests per day. The model supports up to 1 million tokens of input context and offers features like native audio processing, a File API, custom system instructions, and structured data extraction. OpenAI’s GPT-4 Turbo also underwent a major upgrade, integrating multimodal GPT-4 Vision processing directly into the model, enhancing capabilities in writing, math, logical reasoning, and coding. Both models aim to push the boundaries of AI and deliver improved performance across various applications.
Mistral’s Mixtral 8x22B release brought a new mixture-of-experts (MoE) model into the fold, rumored to potentially rival GPT-4 in capabilities. With AI researcher Simon Willison expressing excitement over Mixtral’s potential, the competition among LLMs continues to intensify, with newer models making significant strides in performance. The evolving landscape of AI technology promises more innovation and competition in the future, challenging the established leaders in the field and pushing the boundaries of what is possible with large language models.