Yahoo Scout grounds answers in its own data

Transcript

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

1. 30:23

   are will always have to find that out yeah it's interesting going into that world and talking about your new products especially with scout so where does scout lay in terms of ai search where chat gbt perplexity anthropic even google don't hit it's more i would say from the jump it was more like perplexity. So perplexity is not its own large language model. They were kind of doing meta search on the others, which has happened before. There was a product called Dogpile in the early days of search that was the meta search engine on other people's search engines. Yahoo Scout's totally different. The others are all large language models. So Gemini, ChatGPT, Claude. Ours is a product based on our own data. And there was so much interesting data, which I think another one of our, you know, again, I'll put moat in quotes, because I'm always

2. 31:14

   paranoid about it's not, it's not like you can rest easy with this. But it's an asset. And hundreds of millions of users a month over 30 years, where we're still the third ranked search engine in the US. So we have we have 30 years of search history, even though we're much smaller. It's still a ton of search history, our publisher relationships, which is another way we've taken the company back. Instead of competing with our publishers, we're trying to, you know, make sure we are partnering with them it's our job it's our legacy that we send them traffic yeah and we firmly believe in that it's part of our job in the ecosystem and having been a partner of yahoos at other companies i can attest to how much traffic they were they could send us um so publisher data search data our knowledge graph our user data if we know something about you so we are

3. 31:59

   the way Yahoo Scout works, our AI answer engines, we take all that data, and then we ground it. And then we send it to a lightweight model. At launch, it was Haiku from Anthropic, we use others today. And they process that, send it back to us. And then we, you know, do the rendering that the for, you know, the UI of it. And so what that creates isn't entirely unique results that are only happened because of Yahoo data.

4. 32:26

   and uh well it's interesting because we are also a top 20 publisher whose content has been taken by the large language models without permit permission uh and so you will see the ahu brands show up all the time in citations there um for the most part our content is real time we excel in categories that are real time so we weren't as susceptible to the evergreen content that the large language models kind of photocopied and took with them.

5. 32:56

   So especially in certain categories that are more real time, our results are awesome and totally unique. The other reason why we had to do that is that for 17 years, we were outsourcing search, largely to Microsoft, who we are still great partners with, but that left us in a really vulnerable position for one of our most important products, one of our most important revenue streams. So the ability to fend for ourselves and, you know, own that asset again was just life-changing, you know, for us.

6. 33:27

   Besides the fact that we are, including myself, I spent my first 10 years in search and our old team, you know, we innovated a lot of the most interesting things, especially in structured data search. The things that went beyond 10 blue links is already an area of search I was personally passionate about. So we have our own unique entry there, but it also was just a very strategic asset to own.