#threadcast #CTTPodcast Ep 120 - How #AISummaries project uses AI to summarise Youtube videos using the Hive blockchain as the text storage and data feed and rewards system:
#threadcast #CTTPodcast Ep 120 - How #AISummaries project uses AI to summarise Youtube videos using the Hive blockchain as the text storage and data feed and rewards system:
https://inleo.io/threads/view/mightpossibly/re-leothreads-b3w9gjku?referral=mightpossibly
upvoting all your comments now! Great show!
Was a good one.
Here is something to think about @starkerz ...open full thread.
https://inleo.io/threads/view/taskmaster4450le/re-leothreads-2kgbpnwrd?referral=taskmaster4450le
Quality data is a very important metric, but I would add as well, the fact that its then minable by any other tool without the need of extensive GPU processing needed to check a video that can also later on, be deleted.
But like @mightpossibly, there are many perspectives.
Either way, "text" is a very cheap resource on #hive, and if anyone wants to generate more data, they will consume more RC and that should bring more value in itself, to the chain, just alone on the using the chain.
You're right. That's a perfect system
It's a good thing that you have such a feature that won't allow the same video to be summarised more than once
Would it be possible for AISummaries to do summaries for 3Speak videos too?
It would require a different agent to do that but it could be designed to handle that.
I see, that's good then!
Yes eventually if he gets the time he will be able to build it.
Would be nice to also be able to summarize more things like, this X recording!
Would already have a usecase for @bil.prag that wanted to understand what was talked about in the podcast.
And that brings podcasts more value actually.
here you go sir:
https://inleo.io/@ai-summaries/summary-ctt-podcast-december-31-2024-fmv
It is currently semi-manual labor to summarize x-spaces, due to me having to actually generate the transcript locally first. So no summarize-command yet unfortunately. But feel free to tag me if you find any interesting ones
Really appreciate that.
He does summarize individual spaces.
Really =)
Need to try that...
https://x.com/i/spaces/1PlKQbPMVWdGE
!summarize this space from X mate! - testing...
The bot only works with youtube links, but @mightpossibly does summarize X spaces sometimes. Just tag him.
Yeah I know now... he did it manually for what I was trying to get.
No valid YouTube URL found.
That problem of youtube deleting videos, could probably be solved by doing an evaluation of the video and checking for similar parts. But this would also be a challenge because people that use parts of others videos, could introduce false positives...
But a nice thing to have a look indeed. Having a tool like that for #hive would have a lot of value.
Joel Valenzuela recently had an interview with Julian Chandra of Odysee, a project that decided to move to Arweave blockchain
We should perhaps look at what they are doing to see what Hive Blockchain's differentiator is.
Here's a link: https://arweave.org/
Hive's major advantage is no direct transaction fees.
Most other chains have direct transaction fees that cause an issue for data storage.
hmm didn't realized that caused data storage. Why is that?
That's the best deal on the planet currently
Here is the value of the Hive blockchain:
The ability to provide up to date data....this is where social media apps excel.
Scrapping data for Ai training is one thing; another is to have the models updated outside training.
This is what vector databases tied to social media does.
Thinking through this project - what if any aspect will storage of AI generated data have on the network, and when if ever would the network need additional storage with increased writes to the chain?
It's text... so not a real impact... text is very well compressed now.
This will come into focus when LeoAI eventually rolls out.
The data is the first step to get it posted but it also requires utilizing it.
We were at about 600 posts (comments) in mid November. Here is where it is now.
Holy cow! Didn't expect to be so many. Is there a graph made by someone about who's requesting things... would be a good way to keep an eye about usage.
Not yet, but it's in the works
I just keep an eye on it each day.
Whoa! That's a huge growth in a little more than a month!
Doing around 5K threads per day with it.
If we get a few more people joining in, we can then increase it.
Impressive! Will try to check it out after the new year :D
Hi, @taskmaster4450le,
This post has been voted on by @darkcloaks because you are an active member of the Darkcloaks gaming community.
Get started with Darkcloaks today, and follow us on Inleo for the latest updates.
And this is with just a handful of people. Imagine if we were 20, 50 or even 100 people summarizing videos
Hi, @mightpossibly,
This post has been voted on by @darkcloaks because you are an active member of the Darkcloaks gaming community.
Get started with Darkcloaks today, and follow us on Inleo for the latest updates.
Yep, the gab seems quite yuge
Micropayments are going to be crucial for AI agents.
The cost for a month of access to to this tool is a 5 $HBD subscription. Contact @mightpossibly for questions or more information.
Thanks for the info. That's pretty affordable.
So it's already possible to do AIsummaries for 3Speak vids, it just needs some more functionality features from the 3Speak network. I suppose they will work on that soon😉
Briefly how well does HIVE compare with other chains for storing large volumes of textual data?
I would say a lot... but I don't have any numbers... would need to have a look to show the differences.
I'm do not know how it compares to other chains, but what I can say is that HIVE is really good at it. I am not aware of any other chains that does this this well (or at all)
That would massive! YouTube people do rely a lot on trying to use the "summaries" to create their video timelines. So this could actually create some demand if people find use for it.
Auto-summarising videos from a favorite video creator would be an amazing feature.
I agree! Initially it will probably be more like semi-auto, but yeah. Maybe a bit further down the road
The AISummaries are adding more text data being stored on Hive, the more resources LeoAI will draw from when it launches :)
That is true. It will be up to date based upon what was posted as opposed to being up to the last training date.
Enjoyed a lot this one... great potential on these SUPER active calls. Crazy to think about it, if it had like 1000x more people in these threads/podcasts...
It will happen!
@mightpossibly does currently see any issues with the likely data writes to the Hive Block chain when asked if the @spknetwork would conceivably be need to cover overspill during high utilization.
A good way to somehow decentivize this would be to expose the account using the service. At least reputation account would never do it...
Another way, would be, to require some sort of delegation in order to use the service. To avoid "ghost accounts" doing lots of request without any responsability.
right. Are we talking about duplicate summaries here? Both me (and the blockchain) is keeping track of all summaries, so accountablity/transparency for who summarizes what is already present. Should be even more transparent with the planned front-end and leaderboard.
I'd be interested to hear more about this delegation idea, not quite sure what you're referring to there
Nah, it was the fact that "idiots" might want to attack the chain by scanning for crap videos and flood the command. And how to decentivize doing that...
Right.. Yes indeed, how could we prevent that. It is possible to blacklist accounts of course, but this would quickly become difficult if the userbase increased. Do you think the current 5HBD fee is at least partially contributing to achieveing this?
5 HBD will be more than sufficient I would say. I would start lower to create some adoption and see when the "problem" starts appearing.
Nice!
Instead of making the service free which might attract scripting to do attacks, add a "very small" (suggestion) Hive delegation to an account that you can use for curation. This avoids decentivizes attacks from multiple accounts as they would need to power delegate hive in addition, to use the command.
Then you can even say, that if you delegate 5 HP, you can do 1 command per day, 10 HP, 2 per day, etc... (can be other ratios, just exemplifying).
Obviously this can come later as now raising awareness is probably more important, I am guessing.
Hm I don't think we discussed making it free? I feel like the subscription model is beneficial in several ways, as it both adds utility to HBD and makes it less tempting for spammers/attackers to misuse it. But I'm definitely open to it being things that I haven't thought about yet.
The case of subscribers using scripts to spam the service is somethign I have thought about, and would likely require me to implement some sort of upper limit of daily allowed requests. I'm intriguged by the idea of such a system, but I'm also a bit afraid a system like this would make it unneccessarily complex to use the service
Ups, ignore the free part. I was confusing something in my head.
But on the other hand, it would be important to think about users in general and how confortable they feel with using inleo to pay for a service. Hence why I was referring to the delegation option, which can be also in LEO...
But just to explain, the use of delegations could be an option for longer periods of time as an "accumulator" of tokens that then can be spent by the user at his own leasure.
This in my view is a more attractive way of payment because of its trustless nature. But, obviously spending HBD is more attractive as a financial model to support a service.
Ah yes! This is not a bad idea. You got my mind going. Probably not gonna do this just right now, but definitely gonna keep in in mind for the future - either for this or something else.
Regarding not using INLEO to pay for the service. How about offering an alternative solution: Simply transfer 5 HBD to @ai-summaries, no memos or anything? In addition to the existing option that is.
Hi, @mightpossibly,
This post has been voted on by @darkcloaks because you are an active member of the Darkcloaks gaming community.
Get started with Darkcloaks today, and follow us on Inleo for the latest updates.
I have used the tool already... quite nice!
It might make sense to make an explicit funding post once a week , where people can can come in and upvote it. Presumably it would be supported by by curation services.
Good suggestion! I will definitely consider this
My #hive node is around 507G (not counting history data that is generated from this data) of chain data, in 8 years? Nothing... any disk nowadays can store this, even NVMe's
😎
That's interesting to know! And my thoughts exactly. I was asking in hivedevs discord and they said that a full node with all historical data, 3 tb should be more than enough for the foreseeable future
Yeah, the historical data had a massive improvement with the HAF code. That's on my list for 2025. And it should fit all in a 4TB NVMe... which is crazy to think about!
Excellent.
Data storage moves to data utilizatoin.
Have to build ai agents on top to utilize said data.
Good look storing data on Bitcoin with the transaction fees.
Another point: each youtube url that is posted have the video viewing available on the Hive front end. Over time, this is adding an enormous amount of utility since there is content for people to look at.
AI will help by being able to build a recommend engine.
@cttpodcast theorizes spinning up a breakaway community (a fork of Ecency Code) to document and track the project - and drive traffic to the network for the https://inleo.io/profile/ai-summaries
The future is multi-chain. Interoperability is the name of the game.
#abolishtribalism
I respect and believe in that too
People are creating tokens each day, they are just giving them to Elon and Zuck.
think i am late for this. sounds interesting. will have to get back to it.
You can still catch up :)
Here is how it looks:
!summarize #bitcoin #microstrategy #crypto
Live demonstrations #live! can't really ask for better than that...
Good discussion - properly exchanging value for value - even though @mightpossibly is generating AI data - his intellectual work was needed to create and iterate this system?
@somecallmedan is this a good use-case for DHF funding?
@mightpossibly just in case my tone or context was missed in the original question about network load, I asked because I don't a good mental model for it. My apologies if it sounded like I was implying something negative.
Its good to ask these questions =) people that might have the same worries :D
On the contrary. You're referring to this one, right? I may be missing the point of your question, but I do not see a problem with writing data to Hive. I do find it sensible to avoid summarizing the same content over and over again though. It was perhaps this you was referring to?
https://inleo.io/threads/view/alohaed/re-cttpodcast-q2lg9ham
By the way, I really liked how you kept the threadcast alive during the stream, reporting on what was being discussed as it progressed. Really useful in many ways - for those not listening, those who are half-listening, and also for those coming back reading /listening later after the stream ends
looks like it is close to the end. did you actually did a simple explanation for noobs what is going on here? :)
You listen to the initial call recorded on X...
will replay it, probably right away. was thinking is there a noob explanation i could maybe clip :D
Ai summarized timeline on a clip like this would allow you to quickly go into that part, even if you want to listen to it instead of read (like me, I prefer to listen because I can be doing something else or even driving).
Adding more usecases to the tool. FYI @mightpossibly
Ahh interesting we are getting some history of the development of @mightpossibly AI transcription system.
@theycallmedan with the timely reminder that the Resource Credit system is a rational limit for writes to chain for most users.
It is adding sythentic data to the blockchain.
So subscribe to @mightpossibly and give him 5HBD per month to help offset the cost.
Hello good evening my friend, I think I'm a little late.
Go listen to the recording on X
Ok, thank you
The key is to get ai agents using the chain and people do not even know what chain they are posting to.
100%
You are the best and always supportive.
https://inleo.io/threads/view/khaleelkazi/re-cttpodcast-sicgvw8n?referral=khaleelkazi
This sounds great, but I think it has to be something very stable.
https://inleo.io/threads/view/ifarmgirl/re-cttpodcast-2vmiuq1zv?referral=ifarmgirl
All that information that is shared is stored somewhere, but now I wonder if there is enough space for all that.
Currently all gets recorded on the #hive blockchain as comments. But for sure something to keep an eye for. Just to avoid "useless" things.
@cttpodcast you need #premium to appear in the threadcast stuff... like this...
😎 - I think I have suggested this... but it might have not been me the originator! 😏
Oh! I missed the show. Listening to it now. The AI summarizer is amazing, it summarizes YouTube videos easily and notifies us when we are out of subscription and also if the video was previously summarized.
I wish we could also have for 3speak videos, it will be awesome if this AI summarizer can do that as well.