The company is focusing on lightweight AI models, which are cheaper to run and perform better on personal devices. Microsoft's competitors also have their own small AI models, targeting simpler tasks like document summarization or coding assistance. Phi-3 was trained using a "curriculum" inspired by how children learn from simpler words and sentence structures. Despite its capabilities, Phi-3 cannot match the breadth of a larger language model like GPT-4. However, smaller models like Phi-3 are often more suitable and affordable for companies with smaller internal datasets.
Key takeaways:
- Microsoft has launched Phi-3 Mini, the first of three small AI models, with 3.8 billion parameters and trained on a smaller data set. It is available on Azure, Hugging Face, and Ollama.
- The company plans to release two more models, Phi-3 Small with 7B parameters and Phi-3 Medium with 14B parameters. These parameters indicate the complexity of instructions a model can understand.
- Compared to larger models, smaller AI models like Phi-3 Mini are often cheaper to run and perform better on personal devices. They are also more suitable for companies with smaller internal data sets.
- Microsoft's competitors also have their own small AI models targeting simpler tasks. Examples include Google’s Gemma 2B and 7B, Anthropic’s Claude 3 Haiku, and Meta's Llama 3 8B.