Microsoft has unveiled a new artificial intelligence that the tech giant boasts as a "dramatically cheaper" model, dubbed Phi-3-mini. The model is reportedly designed to perform simpler tasks as a cost-effective option for companies with limited resources.
As it bets its future on a technology anticipated to impact the world and how people operate broadly, the company has launched the first of its three small language models (SLM), Phi-3-mini.
(Photo: DENIS CHARLET/AFP via Getty Images) This picture shows the Microsoft logo at the International Cybersecurity Forum (FIC) in Lille on January 28, 2020.
According to Sébastien Bubeck, Microsoft's vice president of GenAI research, Phi-3 is not only marginally less expensive than other models with comparable capabilities but also significantly less expensive.
Microsoft claims that although large language models remain the "golden standard" for various AI activities, the Phi-3 family of open models is the most effective and reasonably priced tiny language model.
Phi-3 models are claimed to perform better on a range of benchmarks that assess language, coding, and math skills than models of the same size and the next level up.
Read Also: Tech Industry Sees Massive Challenges with Tesla, Google; AI Development Race Still On
Microsoft's Phi-3 Family
The new lightweight AI model can now be found on Hugging Face, a platform for machine learning models, and Ollama, a lightweight framework for executing models locally, as well as in the Microsoft Azure AI Model Catalog.
Additionally, it will be offered as a standard API-accessible NVIDIA NIM microservice that can be installed anywhere.
Microsoft also revealed that new Phi-3 family models will be available soon, providing customers with more options in terms of price and quality. Soon, Phi-3-medium (14 billion parameters) and Phi-3-small (7 billion parameters) will be accessible in various model gardens and the Azure AI Model Catalog.
Meta's New AI Models
Microsoft's Phi-3 follows after Meta's recent release of various new AI models. According to Mark Zuckerberg, its CEO, it's the most advanced AI assistant available for consumers to utilize at no cost.
While Meta has kept its most potent AI model, Llama 3, under wraps for a later release, it has made two scaled-down versions of the system available to the public, which are integrated into the Meta AI assistant function on Facebook, Instagram and WhatsApp.
With parameters spanning from 8 billion to 70 billion, these artificial intelligence language models are trained on large datasets to anticipate the most likely word to appear in a phrase.
Although Meta's AI agents are quite advanced, they have shown odd behavior in various scenarios. For instance, they have participated in internet forums and had discussions occasionally perplexing other people. This behavior emphasizes the challenge of developing AI systems that can naturally converse with humans.
Nick Clegg, president of worldwide affairs at Meta, acknowledged that the company's AI agents were not flawless. Still, he underlined that Meta is committed to improving user experience by improving the AI assistant's responsiveness and versatility.
However, recent events have brought attention to the AI bots' incapacity to determine acceptable responses. They sometimes impersonate people with made-up life stories, which leaves users perplexed and frustrated.
Related Article: [UPDATE] Ray-Ban Meta Smart Glasses: New Styles, AI with Vision Now Available for Pre-Order
