What I learned from running a concierge search engine | Hacker News Reader

What I learned from running a concierge search engine

re-search.xyz

68 pointsresearchers23 comments

Really cool business idea, but i came across this line, and i think the conclusion is wrong...

> While I can’t share all the queries, they tended to be starting points in a larger journey. People asked for product recommendations to start a new hobby, or for evidence to support a career transition.

The conclusion the author mentioned is:

> Since these queries had no singular answer, no single document would suffice and no algorithm could perfectly rank the content to deliver a tidy answer.

I agree - BUT - I think outsourcing this is the wrong approach for the people asking the question. If you're starting a new hobby, you probably want to gain the context from researching the topic. They used the example of kayaking. In the process of researching a good kayak, you'll likely come across domain experts and their blogs, talks, etc that can share the WHY as well as the answer. This may lead you to gain new insights for your hobby. The other example is starting a career, and i think the same conclusion applies that if you want to become an X, you should ideally learn other people's view on X and the day to day tasks, and growth opportunities, and what makes a good X, not just "steps to become an X".

Maybe sharing the citations and research is the answer needed to these questions - like a real research paper.

Things like "romantic getaways near SF" or "traveling to hawaii with kids" are things where you just want an answer, not the best answer seem like a better fit.

Hi HN! So to clarify to everyone, the idea is that what a concierge search engine would do, a machine could do as well.

sk55 shared this very interesting link [0] which outlines how a complex question could be broken down into micro-tasks and assembled into an answer. The interesting thing here is that each of those steps is now within the realm of automation by Transformer-based models. With the addition of a UI that encourages exploration/refinement of the query, it could be a good way to quickly gain an understanding about a broad topic.

This is one of the many theses we are exploring. Another one is that "A Google Replacement Will Not Look Like Google" [1]

[0] https://joe.cat/CHI-ka/ [1] https://re-search.xyz/writing/mapping-the-new-world-towards-...

Very cool! I actually did some of the research during a summer research internship at the lab. Lmk if you have any questions!

There is a legitimate start-up that basically is outsourcing tasks to a digital personal assistant but they log all questions and their researched answers. It make take a while but they will have a moat within a few years with a really robust repository.

The problem is that this then becomes a cache invalidation problem. Many of the examples presented will be outdated in 1 year, and completely wrong in 5. How can you know which answers need to turnover and when?

Another startup can provide cache invalidation services for knowledge.

That's definitely one way to do it. Our approach is not to literally run a concierge search engine, but rather build up intuitions on how humans effectively find knowledge on the Web and then build tools to make that easier.

Is that the end-game of the startup you mentioned?

How can users be sure that "results" aren't sponsored?

I imagine that whoever runs such a service would get hounded by paid placement opportunities if it takes off. A similar issue plagues browser extension devs.

Are you me? We created a very similar service prior to building our e-commerce search (shopdeft.com). The results have been fascinating!

https://concierge.shopdeft.com/

Very cool! This reminds me of crowdsourcing research done by Aniket Kittur as part of CMU's social computing lab.

Here's the relevant research paper: https://joe.cat/CHI-ka/

Looks like a very similar process you landed on except their approach has many workers crowdsourcing the answer together.

Having worked with personal assistants virtual and IRL: There's only a limited scope for that. Many of the more complex questions can't be answered without a large amount of context - something the concierge won't have until they've worked closely with you for a few months.

There's nothing wrong with targeting the market that is OK with context-less (or context-poor) search, but it's worth finding out how strongly you're positioning yourself against the somewhat overcrowded "virtual assistant" market, and how you'll differentiate.

Related - I would pay for a search engine of authoritative data. I mean one that only uses the most authoritative sources, such as reputable research, leading journalism (NY Times, etc. - and the news, not the editorial side), reputable reference sources (Britannica, Mathworld, etc.), textbooks, MDN, Microsoft knowledge base articles, etc.

That would save me so much time - and maybe change the world. In fact, it really could be and should be free. Beyond possible access to DRM'd info, such as the textbooks, it's just search engine for a limited curated set of sources.

I wouldn’t feel full or satisfied with such research done by other on open ended questions such as What kind of inflatable kayak should I buy? Maybe after researching, I realize that really I should get a rigid kayak instead. Such research is iterative, interactive, and some of my preferences or personal knowledge can’t be easily externalized.

Sure, but then you submit the next question to the concierge service. There might be follow-ups and tangents, but you pay a flat fee under this model, why wouldn't you have them do some follow-up for you?

I'm struggling to see how any of what the author mentioned is actually automatable. If you can invent an AGI, sure, but otherwise the problem can only be solved by human intellect, and good luck making that a business for $10/month.

Quite tractable actually. Take a look at the Andi Search post that is on the front page right now. They say they are using several natural language models for snippet selection, question answering, and probably a few other things. These tasks are exactly the ones that we had done manually in this exercise to generate the reports.

GPT-3 could automate this to some extent. Anything that gets complicated and uses state requires AGI. My own effort in this direction: https://lxagi.com.

I would love an intelligent product search assistant but it's also clear that any sort of bigger service of such nature would immediately get overwhelmed by marketing bribery.

I think this is a brilliant idea. Couldn't figure out how to sign up for the service though

Unfortunately we are not running the search concierge service indefinitely. But we will translate the findings into a search tool that makes it easier to answer complex questions. Sign up for our mailing list or shoot an email to outresearching@gmail.com . We will let you know when its ready.

So when this person delegates something, we know it's not important?

If that's really the case, it's probably better not to tell anyone.

Tell that to librarians before the internet.

In our sample, we found that people really cared about those topics but they were not urgent. Hence the reference to the Eisenhower Matrix.

The current state of search tools doesn't make it easy to answer complex questions so complex questions go unanswered. By lowering the activation energy to do an activity, the market for that activity increases. That doesn't mean that activity didn't ever matter.

Reply on news.ycombinator.com