Microsoft has introduced the most recent addition to its Phi household of generative AI fashions.
Known as Phi-4, the mannequin has improved in a number of areas over its predecessors, Microsoft says, significantly in fixing mathematical issues. That is partly a results of higher high quality of coaching information.
Phi-4 is obtainable in very restricted entry as of Thursday evening: solely on Microsoft’s current launch. Azure AI Foundry improvement platform, and just for analysis functions below a Microsoft analysis license settlement.
That is Microsoft’s newest small language mannequin, with a dimension of 14 billion parameters, and competes with different small fashions akin to GPT-4o mini, Gemini 2.0 Flash, and Claude 3.5 Haiku. These AI fashions are usually quicker and cheaper to run, however the efficiency of small language fashions has steadily elevated lately.
On this case, Microsoft attributes Phi-4’s efficiency leap to the usage of “high-quality artificial information units,” together with high-quality information units of human-generated content material and a few unspecified post-training enhancements.
Many AI labs are taking a better have a look at the improvements they’ll make round artificial and post-training information today. Scale AI CEO Alexandr Wang mentioned in a tweet on Thursday that “we have now reached a wall of pre-workout information,” confirming numerous studies on the topic in current weeks.
Notably, Phi-4 is the primary mannequin within the Phi collection to be launched following the departure of Sébastien Bubeck. Bubeck, previously vice chairman of AI at Microsoft and a key determine within the improvement of the corporate’s Phi mannequin. left Microsoft in October will be a part of OpenAI.
