Daniel Svonava pfp

Daniel Svonava

@svonava

59 Following
14 Followers
Daniel Svonava pfp
Daniel Svonava
@svonava
True! Do you know anybody working on this problem specifically? Would love to jam on it!
2 replies
0 recast
1 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
How so? With more content, especially as it gets even more diverse, it will only get more difficult to find great stuff and new creators no?
1 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
Disengaged from the app because content discovery is a mess and it takes too much scrolling to find interesting stuff.
1 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
Is this Sinclair-type BS or actual science? Thanks for the review!
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
My Oura would like a word…
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
My version of “decentralization for efficiency” is making sure our online matrix factorization is properly localized 🫡💅
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
Prediction: ChatGPT will be taken offline due to safety concerns in the next 24h.
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
On a scale from not at all to the first design hire at a startup, how screwed are you? https://i.imgur.com/ZIswbwH.jpg
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
Art can make people cry - no question. But can your own art make you cry? If you’ve been through the creation process and saw behind the curtain? And I don’t mean cry out of frustration when your photoshop crashes ^_^
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
⛏🧱💰
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
This is double-true for any infra/API company. Use hackathons for docs feedback, finding bugs, discovering new use-cases, branding towards devs etc etc
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
I’ve been thinking about this for a while, happy to walk you through our infra that helps you build your personalization engine once you think the time is right :) No strings attached!
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
Yes the general high quality reduces the bar for a good match.. but I’d say that even today on FC, most people will enjoy meeting ~20% of the other users by relevance. Big commit to meet For lunchclub today this is probably <1%. Also, if people get 1-2 bad matches they churn, so you don’t have many shots on the g
1 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
Yes, at scale and based on as many data points as possible. Doable semi-manually for 100s of matches per week if you have some basic tools built around it - breaks for 1000s+. 2 keys to scale: 1. Model that predicts relevance for all pairs of people. 2. Algo that pairs ppl up for that week - “max weight max matchin
1 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
Yep, super annoying, getting the data, lots of natural language processing and feature extraction. We built a tool that makes it easier, would love an intro if you think they would find that type of thing useful :)
1 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
What data do they use? I think that these 3 are a must: 1. Let people express preference using some broad but community specific categories (which cat. do you fall in, which cat. you want to connect with) 2. Let people plug in their socials & use that info to tailor the matches 3. Behavioral data from using the app
1 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
This is very helpful, thank you! cc @dwr.eth a bit of user research 👆
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
So interesting about you getting trained that discover is trash. For this reason, lots of platforms mix discovery into the main feed these days - smuggling in content from folks you don’t follow, to broaden out what you see. How do you feel about that variant? (assuming the underlying discovery engine actually work
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
In a typical community or an event, there are <5% of people you actually want to meet and talk to. This fraction goes lower as communities go mainstream. Source: we have done 3k intros for 20+ professional communities with a demo app we build for our social personalization API.
0 reply
0 recast
0 reaction
Daniel Svonava pfp
Daniel Svonava
@svonava
It’s very hard to scale a social experience without making the discovery actually smart - matching more senior people with other more senior people, matching people with a higher chance to collaborate or enjoy each other because of enough similarity in preferences, yet not totally the same.
2 replies
0 recast
0 reaction