Microsoft launches Phi-4, a new generative AI model, in research preview | TechCrunch
Dec 13, 2024 - techcrunch.com
Microsoft has introduced Phi-4, the latest model in its Phi family of generative AI models, which boasts improvements in math problem-solving due to enhanced training data quality. Phi-4 is a small language model with 14 billion parameters, available for limited research use on Microsoft's Azure AI Foundry platform. It competes with other small models like GPT-4o mini and Claude 3.5 Haiku, offering faster and more cost-effective performance. Microsoft attributes Phi-4's enhanced capabilities to the use of high-quality synthetic datasets and unspecified post-training improvements, reflecting a broader industry focus on synthetic data and post-training innovations.
The launch of Phi-4 follows the departure of Sébastien Bubeck, a key figure in the Phi model development, who left Microsoft for OpenAI. The AI community is increasingly acknowledging the limitations of pre-training data, as noted by Scale AI CEO Alexandr Wang. This development highlights the ongoing evolution and competition in the AI field, with companies striving to enhance model performance through innovative data strategies.
Key takeaways:
```html
Microsoft has launched Phi-4, a new generative AI model in its Phi family, with improved math problem-solving capabilities.
Phi-4 is available on the Azure AI Foundry platform for research purposes under a Microsoft research license agreement.
The model's performance improvements are attributed to high-quality synthetic datasets and unspecified post-training enhancements.
Phi-4 is the first model in the Phi series released after the departure of Sébastien Bubeck, a key figure in its development.
