Simon Willison notes that Microsoft's new models, MAI-Thinking-1 and MAI-Code-1-Flash, mark a departure from the trend of ever-larger LLMs, focusing instead on efficiency and cost-effectiveness. This move could signal a broader industry shift toward optimizing smaller models for specific tasks, potentially democratizing access to AI tools. However, Microsoft's claims of superiority over larger models remain unverified, raising questions about whether these models can deliver on their promises.
The emphasis on "clean and appropriately licensed" data is notable, but without transparency on what this entails, it’s unclear how these models differ from others in terms of ethical training practices. The real test will be in their adoption and performance among developers and enterprises.