Microsoft’s New Love
First came large language models (LLMs) that led to the formation of ChatGPT, built on GPT-3.5 and trained on 175B parameters. Though effective, LLMs are expensive to train, run and can be challenging to customise for specific tasks. Enter small language models (SLMs), which are more efficient to train, deploy and also more accurate. Additionally, they can also run on local infrastructure without resorting to GPU-rich third parties.
Realising the potential of SLMs, enterprises are rushing to develop new small language models.
At Ignite 2023, tech giant Microsoft released the Phi small language model series, termed Phi-2. Phi-2 boasts an impressive 2.7 billion parameters and showcases top-tier performance across benchmark criteria, excelling in areas like common sense, language comprehension, and logical reasoning.
In a blog post, Microsoft said that with the right fine-tuning and customisation, these SLMs can be incredibly powerful tools for applications both on cloud and on the edge. Phi-2 is also available to enterprises in the Azure AI catalogue.
Another interesting fact is that Microsoft claims Phi-2 is open source, which makes it a direct competitor to the LLaMA series of models. Earlier this year, Microsoft claimed that Phi-1.5, which has 1.3 billion parameters, outperformed LLaMA 2’s 7-billion parameters model on several benchmarks.
Recently, Microsoft has been extremely active in the open source space. In June, the tech giant released Orca, an open source AI model designed to learn by emulating the reasoning of larger AI models like GPT-4. The model has 13 billion parameters and is smaller than large models like GPT-4 or GPT-3.5, but it is tailored for specific use cases.
While the discussion about open source continues, experts have pointed out that Microsoft’s Phi-2 is not open source in the real sense as the licence revealed that the model is for ‘research purposes only’ for now.
For Microsoft to replicate LLaMA's success with Phi-2, it must consider making the model accessible for commercial use.
Read the full story here.
Lentra Equips Banks with AI
Since the advancement of AI, the digital lending industry in India has been trying to make the most of it. It is adopting ML models that understand data and decipher the best outcome from it. Lentra AI, a Bangalore-based platform, has been empowering major banks, including HDFC, Standard Chartered, Federal Bank, and many more in India.
Recommended by LinkedIn
The platform uses a consortium of models where the models themselves are orchestrated using elaborate business logic, which makes it slightly more complex than just directly using an XGBoost. Besides, it takes a nuanced approach when it comes to selecting the right borrowers, considering factors like profession, demography and other factors.
Read the full story here.
India Needs AI Skill Revolution
For India to thrive in the AI landscape, sowing an innovation seed in its youth is paramount. A national AI skill development program is urgently needed, requiring substantial government investment. Infosys founder NR Narayana Murthy suggests an annual investment of $1 billion for two decades, accelerating the impact of the National Education Policy (NEP).
To bridge the research-production gap, Murthy advocates recruiting 10,000 retired STEM teachers for a $1 billion/year 'Train the Teacher' initiative. Recognizing the importance of AI in education, initiatives with companies like Salesforce, Adobe, IBM, and tech giants underscore the government's commitment to upskilling students in AI.
Read the full story here.
Microsoft Welcomes All
Microsoft's unwavering commitment to Azure shines as it forges diverse partnerships, including with open-source players like Meta and Mistral AI. This extends to both language model enablers, such as LangChain, and investments in Inflection AI and Hugging Face.
Despite its LLM focus, Microsoft doesn't limit itself. The recent Phi-2 announcement and strategic collaborations with AMD, Intel, and NVIDIA for advanced GPUs underscore its broader Azure-centric approach. The introduction of in-house Azure Cobalt CPU and Maia AI accelerator chip reflects Microsoft's dedicated push to enhance Azure's capabilities, making it the tech giant's true cloud love amid competitive landscapes.
Read the full story here.