Can Perplexity Improve Search Experience with LLMs?

Will Gen AI really improve Search?

Hey Everyone,

The idea that Neeva, Perplexity, Bing AI were going to challenge Google was just nuts. While Neeva was acquired by Snowflake back in May, 2023, and Bing AI sort of crashed and burned in its attempt to take marketshare from Google, Perplexity either needs substantial improvements, more funding and or to have an exit of its own.

Suffice to say that I’m not a huge believer. Its CEO likes to ramble on LinkedIn, and I wish I could say I felt more substance in the future of the product. But perhaps I’m wrong?

Recently, Generative AI startup Perplexity AI has introduced two new large language models (LLMs) that offer real-time access to the internet and information online. Perplexity's new PPLX models leverage the internet for real-time data.

Listen I like You.com and Duckduckgo, but Perplexity isn’t among my favorites.

The pplx-7b-online and pplx-70b-online models are publicly accessible via the Perplexity API and Labs web interface. Now they have partnered with Amazon.

They now leverage Llama-2 and Mistral and have found ways to optimize them. Perplexity’s online LLMs can tap into the latest information from the internet to generate responses, making them uniquely capable of answering queries that depend on recent events or data.

Perplexity is likely to get more funding soon to see if they can really achieve a Generative A.I. version of search that’s viable and competitive with Bing, Google and others.

The startup is itself founded by former Google AI researchers Andy Konwinski, Aravind Srinivas, Denis Yarats, and Johnny Ho. Best case scenario for me is someone like Databricks, Amazon or Salesforce acquires them.

Their last funding round was in April, 2023 so they are very due for more funding, likely around a $500 million valuation, though this could end up being higher.

An “Answer Engine” not a “Search Engine”

Given how Google Gemini is likely to re-frame search, I’m not sure Perplexity’s model is that viable but in a gist, instead of typing keywords and sorting through a list of links, users can pose their questions directly in a conversational way and receive concise, accurate answers backed up by a list of sources.

I have not personally found the citations to be very high quality and directly related to my specific question though, so I’m not sure how viable this product is going to be.

Nor am I clear on how they would make money.

Still the way they are using RAG is interesting. Accessible through the pplx-api and Perplexity AI’s LLM playground, these models represent a novel approach in the realm of AI-driven information retrieval. The company is trying to innovate and that is important and there simply aren’t and haven’t been many search companies willing to do so. Apple might be able to undertake it seriously one day, perhaps as soon as 2025, that is make a viable alternative to Google. But I’m not holding my breath with hope.

Get started with pplx-api here.

Perplexity’s proprietary search infrastructure spanning millions of web pages. The system extracts text and data to augment the models with the latest information for time-sensitive queries.

I hope I am wrong, I do hope Perplexity can build incredible new Generative AI based queries and retrieval systems.

Origin Story

The new models continue Perplexity’s ongoing expansion of its portfolio, including its mobile app and browser extensions.

Familiar evangelists are hyping it up. Generative AI search bots offer a fundamentally different approach to the task of finding content on the internet. The ones that can take videos information and make sense of them have a lot of potential I think. There’s also a lot of untapped data in podcasts and YouTube videos that aren’t very well retrieved around niche queries.

Google’s search experience has gotten noticeably weaker in recent years. ChatGPT isn’t a viable alternative.

Whereas Google works by matching queries with relevant websites found and indexed by its crawlers and ranking them on their popularity and other metrics, Perplexity leverages the capabilities of an advanced large language model to understand the user’s query and its context. This approach has a long ways to go!

Who is Backing Perplexity?

Oddly the backers are mostly so far from major BigTech firms including Yann LeCun, the chief AI scientist for Meta Platforms Inc. Its backers also include at least six current and former AI researchers at Google and sister company DeepMind, including Jeff Dean, Google’s senior vice president for research and AI, who was an angel investor in the company in its early stages, according to data from research firm PitchBook. 

It appears that individuals at Google itself are the main preseed and Series A backers! Among its investors are a striking number of Google’s own AI researchers. So this for me disqualifies this as being a real startup in a sense, the CEO does have a history at DeepMind as a Research intern and even OpenAI according to his LinkedIn profile.

Google is already fated to invest heavily in Anthropic, Character.AI and others. Generative A.I. is consolidating too fast already to have the commercial freedom to innovate a way that will be independent from the magnificent seven, which makes the Gen AI hype train a bit worrisome for the future of the internet. It’s actually a bit surprising and perplexing.

Angels in High Places - AI Bro Support

• Ashish Vaswani, one of the inventors of so-called Transformer models that inspired OpenAI’s GPT-4

• Jeff Dean, lead researcher at Google LLC’s AI unit

• OpenAI co-founder Andrej Karpathy

• Early AI angel investors Nat Friedman and Elad Gil.

So it’s backed by Silicon Valley AI and Angel royalty.

Open-Source Innovation

Still it’s pretty damn impressive that Perplexity is releasing its own AI large language models (LLMs) — pplx-7b-online and pplx-70b-online, named for their parameter sizes, 7 billion and 70 billion respectively.

How they are leveraging Llama-2, Mistral and Amazon is super interesting.

This startup will likely need hundreds of millions of dollars of funding to even build a viable prototype to how Search works today. Can it reach that point?

Join their growing Discord Community!

In addition, Perplexity employs various techniques to maximize factual accuracy and minimize the generation of false information.

Their LLMs are fine-tuned and augmented versions of the open source mistral-7b and llama2-70b models from Mistral and Meta.

Conclusion

If you consider their actual budget in 2023, they have done amazing even to still be around.

How Gemini impacts Google Search will be something to watch, even as Microsoft’s flurry of Copilots might not improve search substantially at all.

Srinivas explained that LLMs will eventually change the way people interact with their computers to find and consume information.

Perplexity’s new LLMs are notable because, in addition to being available for other organizations to use and build their own apps upon through Perplexity’s API (application programming interface), they also aim to offer “helpful, factual, and up-to-date information”. But how many organizations will want to use their API? I’m not so clear on this.

Perplexity is actively building their own web index to make search results even more accurate and relevant. This could shake up the search engine market by offering better ways to find information. With a bit of luck and more funding, they might find a business model that enables them to actually survive.

I cannot say the same for the UK’s Stability A.I. that has a leadership and funding crisis. Because Search Advertising is so lucrative, even a Moonshot like Perplexity is potentially a very valuable startup. Because it’s so rare for anyone to take on Google, the AI community is betting on this startup to at least translate RAG and LLMs into a new kind of search experience.