🤖 Perplexity: Revolutionizing Search with Generative AI
Perplexity is a company that is reimagining the future of search by taking us from 10 blue links to personalized answers that cut through the noise and get to exactly what we want. Their goal is to be the world's leading conversational answer engine that directly answers your questions with references provided to you in the form of citations. In this article, we will explore how Perplexity is leveraging the latest SageMaker innovations to train and deploy their own models, and how they are using generative AI to revolutionize search.
🤖 What is Perplexity?
Perplexity is an interactive search companion that starts with a general question that you had in your mind, digs deeper to clarify your needs, and after a few interactions with you, gives you a great answer. Their Copilot is the first global publicly deployed example of generative user interfaces that reduces the need for prompt engineering. Perplexity is not just a wrapper on top of closed proprietary large language model APIs. Instead, they orchestrate several different models in one single product, including those that they've trained themselves.
🤖 How Perplexity is using AWS to improve their product?
Perplexity started off by testing frontier models like Anthropic's Claude 2 on AWS Bedrock. Bedrock provides cutting-edge inference for these frontier models. This helped them to quickly test and deploy Claude 2 to improve their general question-answering capabilities by providing more natural-sounding answers. Claude 2 has also helped inject new capabilities into Perplexity's product like the ability to upload multiple large files and ask questions about their contents, helping them to be the leading research assistant there is in the market.
Perplexity built on top of open source models like Llama 2 and Mistral and fine-tuned them to be accurate and live with no knowledge cutoff by grounding them with web search data using cutting-edge RAC. They worked with the AWS Startups Team on an Amazon SageMaker HyperPod POC. SageMaker HyperPod makes it easier to debug large model training and handle distributed capacity efficiently. They obtained AWS EC2 p4de capacity for training. This enabled them to fine-tune state-of-the-art open source models like Llama 2 and Mistral and once they moved to HyperPod and enable AWS Elastic Fabric Adapter, they observed a significant increase in the training throughput by a factor of 2x.
AWS has also helped Perplexity with customized service to support their inferencing needs, especially on pf4 and PFI instances, and this helped them to build top of the market APIs for their open source models and their in-house models that have been fine-tuned for helpfulness and accuracy.
🤖 The Future of Search
Generative AI is still in its nascent stages and Perplexity thinks they are at the beginning of what's gonna be a glorious revolution for all of us where the biggest winners are gonna be the consumers of the technology, where we get plenty of choices, great new product experiences, and competitive pricing. Perplexity is closing the research-to-decision-to-action loop even further and they plan to get all their users to a point where we all take this for granted in the years to come. This is disruption and innovation at its prime.
Perplexity strives to be the Earth's most knowledge-centric company and they are glad to be working with AWS so that no one ever needs to go back to the 10 blue links search engine.
🤖 Pros and Cons
Pros:
- Perplexity is revolutionizing search with generative AI.
- Their Copilot is the first global publicly deployed example of generative user interfaces that reduces the need for prompt engineering.
- They orchestrate several different models in one single product, including those that they've trained themselves.
- They fine-tune state-of-the-art open source models like Llama 2 and Mistral to be accurate and live with no knowledge cutoff by grounding them with web search data using cutting-edge RAC.
- They obtained AWS EC2 p4de capacity for training and observed a significant increase in the training throughput by a factor of 2x.
- They have built top of the market APIs for their open source models and their in-house models that have been fine-tuned for helpfulness and accuracy.
Cons:
- Generative AI is still in its nascent stages and there is a lot of room for improvement.
🤖 Highlights
- Perplexity is reimagining the future of search by taking us from 10 blue links to personalized answers that cut through the noise and get to exactly what we want.
- Their Copilot is the first global publicly deployed example of generative user interfaces that reduces the need for prompt engineering.
- They orchestrate several different models in one single product, including those that they've trained themselves.
- They fine-tune state-of-the-art open source models like Llama 2 and Mistral to be accurate and live with no knowledge cutoff by grounding them with web search data using cutting-edge RAC.
- They obtained AWS EC2 p4de capacity for training and observed a significant increase in the training throughput by a factor of 2x.
- They have built top of the market APIs for their open source models and their in-house models that have been fine-tuned for helpfulness and accuracy.
🤖 FAQ
Q: What is Perplexity?
A: Perplexity is an interactive search companion that starts with a general question that you had in your mind, digs deeper to clarify your needs, and after a few interactions with you, gives you a great answer.
Q: How is Perplexity using AWS to improve their product?
A: Perplexity is using AWS to test and deploy frontier models like Anthropic's Claude 2, fine-tune state-of-the-art open source models like Llama 2 and Mistral, and build top of the market APIs for their open source models and their in-house models that have been fine-tuned for helpfulness and accuracy.
Q: What is generative AI?
A: Generative AI is a type of artificial intelligence that is capable of generating new content, such as text, images, and videos, based on patterns it has learned from existing data.
