Meta has released the Llama3 AI model, a chatbot assistant with 70 billion parameters that is free for public use. The Llama3 model has shown promising results in tasks like coding but falls short in math tasks compared to other models, with anticipation for a larger 400-billion parameter model in the future.
Meta has released the Llama3 AI model, which is an AI chatbot assistant similar to GPT-4 and has impressed users with its performance. The Llama3 model is open and free for everyone to try, and it has shown promising results in various tasks such as coding and scientific tests. The model’s 70 billion parameters allow it to perform well in tasks like coding, with an 82% success rate in the HumanEval benchmark. However, its performance in math tasks is not as strong compared to other models like GPT-4.
There is anticipation for a larger 400-billion parameter model from Meta, which is currently in training and expected to show significant improvements over the 70 billion parameter model. Despite the benchmarks and success rates, it’s important to consider the validity and significance of these tests, especially when they exceed 80-85% success rates. The GPQA scientific test is highlighted as a good benchmark for evaluating AI models.
The Llama3 model has been compared favorably to earlier versions of GPT-4 on the Arena leaderboard, showcasing its performance in English language tasks. Additionally, Google DeepMind’s Gemini 1.5 Pro AI model is also making strides, particularly excelling in tasks involving longer prompts. Both of these AI models are available for free use, providing an opportunity for users to experience advanced AI assistants without cost.
Users can try the Llama3 AI model by visiting Meta’s website, downloading it for use, or accessing it on websites that host the model. The rapid improvement of AI assistants and the resurgence of open-source models are highlighted as key insights from this development. The availability of high-performing AI models for free use signifies a significant advancement in AI accessibility for users worldwide. Overall, the progress in AI technology and the availability of open-source models present exciting opportunities for users to experience cutting-edge AI assistance.