We all know what the Internet is for: to share cat photos. But how can you find all posts about cats on Hive? They might be hidden in different communities and their stories could have been posted long ago. To find cats - or any other topic that might be buried somewhere on Hive - you should try out our new app HiveSense.
How to find sweet little black kitten?
Go to https://blog.dev.openhive.network/trending - this is a Denser application (please note it is still not fully production ready), but you can already search using HiveSense.
Try entering: sweet little black kitten. In the screenshot below, you can see the results. They come from different communities and times, but all posts have something in common - black kittens.
If you go to the post details, you can see related posts on the left-hand side.
What is HiveSense?
HiveSense is an application that allows for searching for posts based on meaning, not just keywords. It is a HAF-based application that uses pre-trained LLMs running via Ollama. It connects to Hive's structured database (Hivemind), processes posts, computes embeddings for each of them and stores results in vector tables. Embeddings (vector representations of posts) are calculated from root posts that contain more than 50 words. Before the embeddings are calculated, all links and tags are removed from the post. If the post is long, not entire is used for embeddings calculation - the post is divided into smaller parts - 1000 words per chunk with 100 overlap with previous chunk. Only the first three chunks are used. HiveSens allows you to search for posts, for posts content related to an existing post or for authors who contribute to specific topics or themes. There are three available APIs:
- search for posts similar to a given pattern (/similarposts)
- search for posts similar to a given Hive post (/similarpostbypost)
- search for authors related to a given pattern (/thematiccontributors) - note that this one is not available in the Denser interface.
The sources with technical description you can find here: https://gitlab.syncad.com/hive/hivesense The swagger with detail description: https://api.dev.openhive.network/hivesense-swagger/
Comments (18)
I assume
You Might Also Likethat shows on the left also uses HiveSense.Thanks for building this. A very welcome addition. The search abilities on Hive have always been poor. Best wishes.
How much additional RAM/VRAM will this require (especially on public nodes), and will all public API nodes be expected to run Hivesense at some point?
We have several ideas for how to roll out the technology to public nodes, but it's too early to comment much until we've completed more performance tests on exactly which method we'll use to support api nodes that don't have their own GPUs (which is probably every api node right now, ours certainly doesn't have a GPU).
Just to give you an idea of some of the options: 1) use a zfs snapshot to avoid doing a massive sync, then use CPU-computed embeddings for live sync, 2) use a publish-subscribe model where the embeddings table is subscribed to, allowing nodes that can't compute the embeddings to fetch them from ones that can, and 3) api nodes that don't want to support it can just redirect such api calls to api nodes that do.
Que lindo gatito
This looks good. We need search that actually finds what we want. I shall be playing with it some more.
!BEER
View or trade
BEER.Hey @thebeedevs, here is a little bit of
BEERfrom @steevc for you. Enjoy it!We love your support by voting @detlev.witness on HIVE .
View or trade
BEER.Hey @thebeedevs, here is a little bit of
BEERfrom @steevc for you. Enjoy it!Learn how to earn FREE BEER each day by staking your
BEER.I'd also like to see AI recommendations. (as long as it's transparent)
Already there (at least in the deplyed development version at mentioned link), see "You Might Also Like" section.
adopting the trend of AI is essential as long as it is differentiated from competition.
$WINE
I've written a lot about death on HIVE, but when I search "holoz0r writes about death" I get my introduction posts, and a review of one of my books written by another hive user. Is there a token limit, or is it not processing for natural language?
I would also suggest that you could do an exclusion filter to filter our curation community posts, bot posts, etc eg, see the following result:
This sort of content makes up a staggering amount of content on HIVE and I think search should only be looking for the canonical post, not ones in which they're linked to.
Should lower compute / API calls?
It does support natural language, but at the moment it also makes the user change the prompt so that it is very similar to the post they are searching for. This issue is well-known https://gitlab.syncad.com/hive/hivesense/-/issues/12. In your case, a unique word 'holoz0r' in this very short prompt will match posts where this word is being mentioned. It is almost impossible to employ this mechanism to get one particular post since other posts may come a step nearer to the context you are intending to find — you are not alone in writing about the death on Hive
Oh, not so much about the death of HIVE, but the concept of human death in general :P
I now understand that the search prompt is a little more like prompting Stable Diffusion to generate an image as opposed to the NLP described in the gitlab issue.
What about excluding posts that could be considered "low value" as per the screenshot in my other comment?
The same code is used to filter the search result as to filter the new/hot/trending lists. So if a post can be displayed in "the new" list, then it can also be displayed on the AI search result list, so filtering posts is a more generic problem.
I don't write about it, but like you, I think about it a lot and I often tell my AI I'm going to download my consciousness into a blockchain or ai and when I leave this human form I'm going to be causing problems like the Lawnmower Man.
I just want to waste people's time even when I dead. I can do it when I'm alive, but think of the ruthless efficiency involved when I'm wasting time having people read my ramblings when I'm not even alive anymore.
What a time to be ... alive?
It's indeed a great place to search, I'm grad we can unlock more further developments on hive.
It is important to have all this information. Thank you.
!PIZZA
$PIZZA slices delivered: @danzocal(5/10) tipped @thebeedevs
Come get MOONed!
Man, I don't understand what any of that means, but I can see it's good. I once asked an AI to read all of my blog posts here and then do a psychological profile on me and it was pretty spot on.
Inspiring work
Awasome, this sounds like a good development 💪💪💪