MS launches small language model 'Pae-3'... Cost reduced to one-tenth

Microsoft (MS) announced on the 23rd (local time) that it will launch a small language model (SLM) called 'Phi-3 Mini'.

Phi-3 Mini is the lightest model among the three small models that MS plans to release.

MS released Phi-2 last December. MS explained that Phi-3 performs better than the previous version and can provide responses at the level of a model 10 times larger.

SLMs operate on personal devices such as smartphones and laptops and are more cost-effective compared to large language models (LLMs). MS explained that SLMs are designed to perform simple tasks, making them easier for companies with limited resources to use.

S?bastien Bubeck, Vice President of Generative AI Research at MS, said, "Phi-3 is not just slightly but significantly cheaper," adding, "Compared to other models with similar functions, the cost is about one-tenth."

The Phi-3 Mini has 3.8 billion parameters. It is smaller than the upcoming Phi-3 Small with 7 billion parameters and Phi-3 Medium with 14 billion parameters. Parameters indicate how many complex instructions an AI model can understand.

Eric Boyd, Vice President of MS Azure AI Platform, said, "Phi-3 was built based on what was learned from previous versions," adding, "Phi-1 focused on coding, Phi-2 started learning reasoning, but Phi-3 is more proficient in both coding and reasoning."

With the launch of Phi-3 Mini, MS is expected to intensify competition in SLMs in addition to LLMs within the US big tech industry.

Google released Gemma 2B and 7B in February, which are useful for simple chatbots and language-related tasks. Meta launched 'Llama 3' on the 18th, introducing a large model with 70 billion parameters and a small model with 8 billion parameters used for chatbots and coding support. AI startup Anthropic released Claude 3 Haiku, which can quickly summarize research papers with graphs.