thing that uses a dimensionality reduction to lay out some dots and put labels on em, so I can see all the stuff organized by category. invited you to a recent one I was playing with on gh. i’ve had some fun with nomic but have some intense persistent frustrations that make me less inclined to use it (it’s missing things I find obvious).
i was accidentally ambiguous in my previous comment—I used to do what op describes, it’s how I got the videos I’ve shared. but, I didn’t feel rushed to correct it, because I actually did also do vector embed then dimred on YT videos, and that was also part of how I found cool videos to share here. it just happens to not be what i intended to describe.
vector embed UIs are imo the best when they’re persistent, which is achieved with a parametric embedding.
I’ve been thinking LW really really needs one so I’ve started tinkering to make it happen, but as usual I’m a bit slower than ideal-me due to distractability and suboptimal prioritization.
I’m surprised that this is particularly practical, and the details here were more of what I meant to be asking about. (Still interested in more detail)
wait can you explain what you mean by vector embedding UI in this context? How did you do that?
thing that uses a dimensionality reduction to lay out some dots and put labels on em, so I can see all the stuff organized by category. invited you to a recent one I was playing with on gh. i’ve had some fun with nomic but have some intense persistent frustrations that make me less inclined to use it (it’s missing things I find obvious).
i was accidentally ambiguous in my previous comment—I used to do what op describes, it’s how I got the videos I’ve shared. but, I didn’t feel rushed to correct it, because I actually did also do vector embed then dimred on YT videos, and that was also part of how I found cool videos to share here. it just happens to not be what i intended to describe.
vector embed UIs are imo the best when they’re persistent, which is achieved with a parametric embedding.
I’ve been thinking LW really really needs one so I’ve started tinkering to make it happen, but as usual I’m a bit slower than ideal-me due to distractability and suboptimal prioritization.
Are the vector embeds, like, part of a browser extension that is tracking everything you click on yourself or what?
That would be cool but would be bad at finding new things.
but like where are you getting the data from? that’s what I was confused by
Downloading it? Normal data getting. List of things to download, automatically download em all. Took some effort to set up
I’m surprised that this is particularly practical, and the details here were more of what I meant to be asking about. (Still interested in more detail)
I’d be interested in a GitHub invite to the repo to check how you’re doing this too.