From: redpointai
In the coming decade, search technology is expected to transform into a system that provides direct answers and agents capable of completing tasks, mimicking conversational interactions with friends [00:00:01]. This evolution moves beyond mere links to comprehensive, concise answers [00:23:25].
Perplexity AI’s Vision for Search
Perplexity AI, a “next-gen search product,” aims to maximize “knowledge velocity” – the rate at which knowledge is acquired [00:15:40]. This contrasts with older platforms like Quora and Wikipedia, which focused on maximizing knowledge itself but lacked speed or personalization [00:15:55]. Perplexity’s goal is to fulfill all human curiosity [00:17:22].
Key Dimensions of a Successful Search Product
Perplexity AI focuses on five critical dimensions to deliver a top-tier product [00:01:52]:
- Accuracy [00:01:34]
- Reliability [00:01:37]
- Latency [00:01:38]
- Delightful User Experience (UX) [00:01:40]
- Iterative Improvement: Constantly enhancing for everyone or through personalization [00:01:42], [00:01:48].
Evolution of Search Results and Interaction
Modern search goes beyond simple text:
- Multimodal Answers: Answers should not be limited to text. Images are “worth a thousand words,” and videos are “worth tens of thousands of words,” making them crucial for accompanying search results [00:02:52], [00:03:00].
- Sharability: Answers should be sharable entities, such as permalinks, so others can benefit without re-asking the question [00:03:09]. This innovation was adopted by competitors like Bing and ChatGPT [00:03:26].
- Follow-up Questions: Search engines should suggest follow-up questions because humans are not always good at articulating their curiosity into precise queries [00:03:30], [00:03:42].
- AI-Assisted Query Formulation (Copilot): Instead of blaming users for poor “prompt engineering,” tools like Perplexity’s Copilot assist in refining questions [00:04:25]. The philosophy is “the user is never wrong” [00:05:27]. The ideal scenario is for AI to magically know when to clarify or dig deeper, without users having to pick a mode [00:31:36], [00:32:01].
- Personalization: Future search will cater to individual depth preferences, such as a high-level overview of black holes versus detailed celebrity gossip, or vice-versa [00:16:00], [00:16:36]. This “fast, high bandwidth access to knowledge” is a core tenet [00:16:38].
Challenges and Strategy in AI-Powered Search
Perplexity AI views search as a “vertical” within the broader context of AI chatbots, and focusing on knowledge and research within that is another vertical [00:17:46], [00:19:01].
Competing with Incumbents
Challengers to dominant search engines like Google often fail by trying to replicate Google’s exact approach (crawler, indexer, ranker) [00:20:12]. Successful challengers, like DuckDuckGo and Brave, found unique positioning (privacy, crypto) and focused on “go-to-market and market positioning and branding” rather than building technology from scratch [00:21:17].
The emergence of powerful models like GPT-4, and their accessibility via APIs, created a rare “moment” where Google was not the leader in AI, providing an opportunity for new players [00:22:46].
The Role of Models and Vertical Integration
Perplexity’s strategy has evolved from using off-the-shelf models to fine-tuning smaller, faster models, and eventually releasing their own open-source models [00:07:27], [00:07:34], [00:07:38]. They advise product-focused companies to start with existing models to achieve product-market fit quickly, rather than immediately investing in model building [00:08:04].
Perplexity aims to be “model agnostic” but retains the option to control its destiny if needed [00:13:44]. The user’s primary concern is getting the best answer, not which model powers it [00:14:06].
Addressing Hallucinations with Retrieval Augmented Generation (RAG)
Perplexity has been highly effective at using RAG to combat hallucinations [00:47:45]. However, solving RAG for web search is different from solving it for internal enterprise search, as indexing, embeddings, and snippet generation vary significantly [00:28:18], [00:28:32], [00:28:35], [00:28:40]. Throwing more information at long-context models can even increase hallucination chances, emphasizing the need for robust retrieval, indexing, and ranking [00:30:07].
Product Development and Growth
Perplexity made a crucial pivot to consumer-facing search when internal tools, designed for self-use, generated unexpected excitement among early users [00:44:56], [00:45:09]. A key virality moment occurred when their Twitter search product summarized public social media activity, leading to users sharing screenshots of their own profiles, akin to ChatGPT’s screenshot virality [00:48:21].
Optimizing for new users is crucial for growth, even over catering to power users who might demand specific features [00:34:55], [00:35:51]. Products that are unintuitive or blast users with too many sign-up modals tend to stifle growth [00:36:34].
The Future of Voice Interfaces
While some envision future search being entirely voice-to-voice, there’s a concern about AI “rambling,” as eyes can read answers faster than ears can hear them [00:23:51], [00:24:12]. Conciseness is key for effective voice interaction [00:24:10].
Future AI Applications and Content Creation
Beyond traditional search, AI will enable new forms of interaction and content creation. The ability to record loved ones’ voices and personalities to interact with them in the future is a personal project aspiration [00:56:03]. Similarly, reducing the marginal cost of creating movies and visualizations through generative AI tools could lead to an abundance of “amazing” content that explains concepts better [00:56:54], [00:57:00]. This also ties into multimodal reasoning and generative user interfaces [00:53:07], [00:52:28].