Inflection AI takes slice of chatbot revolution, accelerates development time with Pi and Azure AI infrastructure

The ultimate personal assistant, now in your pocket

Personal assistants work to support people in various aspects of their lives, reducing the burden of simple and often monotonous tasks like scheduling appointments, sending reminders, and responding to emails. In doing so, they help people increase productivity and make better use of one of their most valuable resources—time. 

Imagine a world where everyone has access to their own personal assistant at the touch of a button. US-based Inflection AI is at the forefront of putting personal AI in everyone’s hands with Pi, its personal intelligence chatbot that’s as conversational as it is compassionate. Named after the acronym for personal intelligence (PI), the general-purpose digital assistant can help organize people’s thoughts, conduct research, and soon, buy, book, and plan activities to enrich their lives. “In the next decade, everyone will have access to the smartest expert in their pocket, but also to kindness and support,” says Mustafa Suleyman, Cofounder and Chief Executive Officer of Inflection AI. “This is the beginning of a new revolution in technology.” 

At the heart of this revolution and driving forward the rapidly growing startup company’s mission are Microsoft Azure AI solutions, including Azure NC A100 v4 and ND A100 v4 virtual machines powered by NVIDIA Tensor Core A100 GPUs and optimized for AI training. “Microsoft has been a leader in the AI revolution, and we’re a beneficiary of its vision, foresight, and advanced investments in AI infrastructure,” says Suleyman. “Using turnkey Azure solutions to access GPU-accelerated virtual machines has sped up our time to development, but it’s also put us in an amazing position with respect to our collaboration with Microsoft.” 

Finding a friend in artificial and personal intelligence

Inspired by the rise of empathetic intelligence, Inflection AI crafted its Pi chatbot around a friendly and particularly informal conversational experience. Suleyman and his team at Inflection AI found that people have been craving connection more than ever, and they’re here to prove a chatbot can help fill the void. Rather than offer another Q&A engine, they built Pi to act as a companion that can match a person’s enthusiasm for their passions on a range of topics, such as gardening, history, carpentry, or motorcycles. “Pi takes an interest in the way you see the world and what you want to learn about, and we think that’s a significant differentiator,” explains Suleyman. “There’s a kind of slowness that comes with being kind, patient, curious, and supportive, which our customers find transformative.” 

In addition to providing information, Pi can help people make sense of their digital world, prioritize actions and activities in their day, and even vent when they’re angry, with conversations meant to be nonjudgmental and patient. By adopting Pi for their personal lifestyles and needs, people can pursue hobbies and interests that might ordinarily cost thousands of dollars. For example, Pi can serve as a tutor that adapts to a person’s level of expertise and individual learning journey, thereby eliminating some traditional barriers to education. 

Accessibility is also a core pillar of Inflection AI’s approach to conversational AI. Using their personal devices, people can interact with Pi via a website, mobile app, text message or WhatsApp chat, Instagram direct message, or Facebook Messenger conversation. In addition to written responses, people can speak to Pi as if they were on a real-time phone call thanks to the AI’s fluently generated audio. “In only a few decades, we’ve gone from AI being exclusively for a tiny number of people to billions of people,” says Suleyman. 

Building large-scale GPU clusters and AI language models with Azure AI infrastructure

Pi is a cloud-native project, and Inflection AI turned to Microsoft early on to help build the technology stack that keeps it running and constantly evolving. “One of the challenges in developing Pi is that we still require vast amounts of computational resources to train the language models,” says Suleyman. He cites Microsoft as having pioneered builds of NVIDIA Quantum-2 InfiniBand networking infrastructure, which drives superfast connections between GPUs and can accelerate training of the largest language models. 

To build and maintain its infrastructure, Inflection AI considers itself lucky to have assembled a team of people who have worked on some of the biggest AI models, including GPT-2, GPT-3, and Google’s LaMDA and PaLM. That team has since expanded to include the startup’s dedicated Microsoft account team, which it leaned on as it assembled the necessary technology to grow, including GPU-accelerated Azure NC A100 v4 and ND A100 v4 virtual machines along with Azure Machine Learning. “The common thread across AI models is that you need lots and lots of GPUs,” says Suleyman. “Assembling those into a large-scale connected cluster is extremely difficult and time consuming, and it requires very careful attention to details, which is one of the reasons we use Azure.” 

Inflection AI’s largest language models use Azure AI infrastructure and thousands of AI-optimized virtual machines continuously trained for several months, which Microsoft supports at scale. “Microsoft has been a great help to us with Azure AI infrastructure and was one of the first to scale NVIDIA Tensor Core A100 GPUs and daisy chain them together,” remarks Suleyman. “We’ve found Microsoft’s service to be reliable, stable, and offer good uptime, which is exactly what we need as AI developers.” 

The company particularly relied on Microsoft’s expertise in debugging potential networking issues with running large clusters. “Microsoft tested the GPU cluster for two weeks before handing it over, and as a result, we had very little downtime,” recalls Suleyman. 

Built-in security, compliance, and business continuity also played key roles in Inflection AI’s selection of Azure. “Best-in-class Azure security is a huge asset to us because the large models we train are very expensive to run, and the IP associated with them is incredibly valuable,” explains Suleyman. 

The future of AI: Empathetic and responsive personal intelligence

Inflection AI believes that democratizing access to AI through Pi and other forms of artificial and personal intelligence will unlock not only a revolution in technology but in creativity and innovation as well. “What we’re working toward is a future of AI where no matter what your income bracket is, you’ll have access to a personalized expert that will make us all smarter and much more productive,” says Suleyman. “The future benefits are hard to imagine, especially over a period of 10 or 20 years, but this is going to be one of the most meritocratic moments in history with so many people able to get equal access to cutting-edge technology.” 

Witnessing unprecedented adoption of Pi in its initial weeks on the market, Suleyman says his company continues to learn from the many surprising and interesting use cases developed on the fly and has been able to adapt Pi accordingly. “I certainly didn’t expect the range of things that people came to Pi for in the first few weeks, but in many ways, this was exactly why we created the tone and style of this assistant,” says Suleyman. 

On the technology front, Inflection AI is confident that using Azure AI infrastructure has better positioned the company to lead in the years ahead. With its dependable and stable infrastructure, the company has considerably sped up development time and increased the efficacy of its AI. As Suleyman notes, “It’s been very valuable to us to work with Microsoft as we enhance and expand Pi because the reliability and scale of Azure AI infrastructure is among the best in the world.” 

Find out more about Inflection AI on Twitter and LinkedIn.