Hi, I'm Charles and I made Ozlo along with my co-founder @michaelrhanson and a team of other smart people here in Palo Alto and Seattle. If you want to try Ozlo sooner, go to our website at signup at http://ozlo.com?vip=producthunt
By the way - Alex Kantrowitz at BuzzFeed has been using Ozlo for the last few weeks. You can see his experiences here: https://www.buzzfeed.com/alexkan...
@bentossell Ozlo is an AI bot that can help you find information and make decisions. We call him personal AI because he works for you - you can tell him about yourself and your preferences, tell him when he is right or wrong, and he will accommodate himself to you.
He's different that most assistant/bots because he actually has his own internal model of the world, built by reading the content from many different apps and publishers. This allows him to have a deeper understanding of the questions you ask him and your own personal preferences.
if you've tried any of the bots recently and found them disappointing, you'll see how Ozlo is a major step forward. You can see Ozlo in action in our opening blog post https://medium.com/teamozlo/intr....
Or signup and install the app on the App Store – we'll let you in as fast as we can. Also happy to answer questions here.
@okito@michaelrhanson Hunted by @johnolilly, beta tested by @sarahtavel... should we read between the lines :-)? Congrats on the launch anyway.
I was wondering: is it US only?
I've also been a tester for Ozlo for a while now. What I really like is that I don't have to fumble between different apps to find what the star ratings from Zagat or Foursquare or Yelp. All that info is easy to get to AND it's done using the conversational syntax that I've gotten used to with other interfaces like Siri, Alexa or Google Now.
Why do you guys believe that AI and the conversational interface is getting so much hype right now and what makes you think it's the right one?
@coreyo Hey Corey! You know it's a lot like what happened on the web - at first people were so excited about the web, they didn't mind having to find all the website they wanted manually and bookmarking them. After a while, there was so much information, you needed something to help you get to the right place –– and that is where Google emerged.
We think we're at a similar point now with phones and apps. People are growing tired of trying to manage their own apps. You need a new starting point that is easier and more direct for the phone. AI and conversation are the most natural fit.
It takes a long time to build an AI though. So although a lot of people are getting it now, I think if you haven't been working on something for a couple of years already, it's going to be hard to build something amazing.
Report
This is cool, i just got access. I love the design and surfacing cards that have more robust info - reviews, articles, etc. Ozlo is also cute :) Curious why you set out not letting Ozlo work conversationally but more structurally?
@dleesta he does both! just tap on his avatar to text with him. The suggestion bubbles are actually just accelerators to save you typing; you can text anything you see there (or similar phrases)
We did it this way to help solve the discoverability problem of conversational interfaces. Behind the scenes Ozlo has a complex system that can sense which suggestions are most relevant to you. It's still early but will improve fast.
Report
@okito gotcha, yes i see that now. are you only supporting place recommendations right now?
@dleesta Ozlo is can become an expert on any topic we choose to feed him, but we've focused so far on places because it's something everyone can relate to and it's technically very challenging. We realized if we could do places well, we could do just about anything.
Report
@okito signed up this morning, can't wait to try out this product! I'd love to hear about ML algorithms you've used to implement this technology. Is your NLP based on RNNs? Or did you start with a human-annotated dataset upon which to build this AI (similar to what x.ai did)?
@alekslyng@okito Ozlo is actually a stack of different models that we use to understand the world and the user's language, and we use different bits of ML depending on what problem we're solving. The NLP model is supervised right now.
Report
@michaelrhanson got it, thanks! So many awesome things happening in this space right now and as a non-technical guy, I've been trying to understand how products like this are brought to life. Seems like you guys have created something really neat - can't wait to try it out.
I've been alpha testing Ozlo for the last week or so around SF, and it's saved me a bunch of time. It's fully replaced using the native Yelp app for me for restaurant recommendations (which the alpha is limited to, but I hear it will expand to much more) - asking Ozlo something like "Great sushi spots around here that are open for lunch" is so much faster than futzing around with Yelp's filtering UI.
@eriktorenberg Great question. Building an AI people will actually trust and rely on everyday is extremely challenging because building the first version is just the beginning. There is a lot that goes into keeping Ozlo fed with new information so that he continues to grow and evolve that we have to do a great job of. It depends a lot on getting good feedback from early users like you!
Let me start my (additional) feedback with THIS SHIT IS HARD, building a neural network and allowing accurate decisions to be made is hard. I concede to that. This is just user experience feedback.
I was playing with Ozlo this evening, and I asked it (in various forms of the same question) "Where can I find the best cheesecake in San Francisco?"
The result I got back was something like "I know lots of good restaurants in San Francisco that serve cheesecake, but some are far away: [a long list of restaurants, like a Pizza restaurant in position #1]"
I actually expected something more like: "based on what I could find, most people would agree that you would find the best cheesecake in San Francisco at either of these three locations: [three locations listed]"
Then a follow on "Do you want the cheesecake today?" if yes: "Darn, the best ones are closed but I recommend you try these three for tonight: " || "Darn, only two of the best locations are open right now"
I like the parts where you guys ask if he got it right or not but perhaps there can be more action there like: "Would you like to order a cheesecake from A || B || C" or "Would you like to make a reservation at A || B || C?"
Questions to fall out the loop are great and useful but I would love if Ozlo could follow my intention. My intention being getting the best Cheesecake San Francisco.
Then as part of its learning if I go with an option it could ask me a question the next day like: "How was the cheesecake? Would you say it's the best in San Francisco?" and learn from my response. Which could be "yes, it certainly was!" or "No, not really but their Key lime is incredible" and so goes the learning.
Your design and the overall brand is so soft, soothing and forgiving that I wouldn't even be mad at it if it got it wrong. I would love to engage it to make it get to know me, as it follows my behaviors. etc.
Point is I think my AI butler / concierge should give me firm recommendations limited to 3 other wise how is it different from search on Yelp I guess?
With all that said, I love Ozlo already and look forward to using it and watching it learn and grow. :)
@mitchellgeere Thanks for such detailed feedback! I totally agree with pretty much every point you make here. – the main thing being that a great AI like this should be able to think about your intention, not just individual searches. We recently introduced a new system inside of Ozlo based on Interactive Fiction technology that is giving Ozlo that ability to do this. So stay tuned; it's going to take some time for him to do this all the time – but what you describe is very close to how we think of it also.
Palua
Palua
Palua
aiden.ai
Palua
Emisare Inc
Palua
Palua
Palua
Palua
Palua
The Freelancer's Journey
Palua
Backchannel
Palua
Daybear
Palua