Real Alpha
Watch modeSubscribed

Lightcone · 2 min · Feb 21, 2025

Treat ambiguity as a product problem, not a user failure

Aravind Srinivas says AI should clarify ambiguous requests instead of blaming users for poor prompts. That principle shapes how Perplexity handles errors and designs its consumer experience.

Aravind Srinivas

Aravind Srinivas is cofounder and CEO of Perplexity.

Held research roles at OpenAI and Google DeepMind before co-founding the AI search engine Perplexity in 2022.

  • DL

    David Lieb, Host · General Partner, YC

About this segment

Aravind Srinivas says AI should clarify ambiguous requests instead of blaming users for poor prompts. That principle shapes how Perplexity handles errors and designs its consumer experience.

From “How To Build The Future: Aravind Srinivas”, Feb 21, 2025. Watch the full episode

“Enterprise software is more like second kind, but magical consumer products are more the first kind.”
Aravind Srinivasat 20:13 · Lightcone20:13

Takeaways

  1. When a query is ambiguous, ask the user to clarify.
  2. Do not make people learn to prompt around product shortcomings.
  3. Consumer products should adapt to users.

In this episode

Full episode
  1. 10:32A simpler search system worked as language models improved3:11
  2. 27:18Search needs to help people complete the task, not just find answers3:44

  3. 19:00Treat ambiguity as a product problem, not a user failure2:13

  4. 5 more moments in the full episode
Up next · Big Technology PodcastDeception can undermine every other AI safety test4:34

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 19:06

    of the culture of the early days of google like the things i've learned about larry or about sundar and i see a lot of that in the way that you have built your product like there's a lot of attention to detail feels like you are the primary user of the product yourself like is that a thing that you deliberately tried to do yeah i did i did deliberately try to do it uh one thing that larry says like you know we we uh i keep reminding everyone in our company about it the user is never wrong so even today while testing a new feature um it didn't work uh but there was some ambiguity in the query so the person i was talking to the engineer and saying hey you know this is not good. What else could AI have done here? And you know what the AI should have done? It should have come and clarified to me and asked me, hey, I'm not sure either this or this. Which one did

  2. 19:58

    you actually want? And then I should have clarified. and then it should have gone and done it. Instead of saying, I don't know, that is the user is never wrong principle. The other way of designing products is like, make the user be a better prompt engineer. Blame the user and tell them to be a better prompt engineer. Teach them, educate them, get them to do it the way that the product wants them to do it.

  3. 20:17

    Yeah, exactly. Enterprise software is more like second kind, but magical consumer products are more the first kind. I agree. Like in Google, why should Google handle typos? They needn't have, right? We should have all been great at English. It's like Larry says he was never good at spellings, and that's why. I think the true story is YC partner Paul Bukai, he was just annoyed by it. And he's like, someone should build that.

  4. 20:40

    Yeah, exactly. And spell check, corrector, it's all there. Similarly, auto suggest. Why is it there? Like easier, right? Similarly, cached results. I was even reading somewhere where Larry wanted the homepage to have the simulation of the weather outside your home so that you don't even need to type the weather query. It's just already there. So I was very influenced by that style of design, like including like subtle things, the Chrome search bar, if you've already gone to a site, it's already there.

  5. 21:08

    You just have to click enter after typing the first two letters.