Hey! Thanks for making such a great tool!
I have not forked FOSSBot. After some tweaking, its parameters are sufficient for my needs for now. I do have a list of improvements I might make - but these are quite far down the priority list. If I do fork then I'll consider a pull request if I think my changes are of interest to anybody.
My position is that filtering is a UI level task, not a blockchain task. How that gets implemented is really up to the UI creators and their own philosophy on how they'd like their UI to operate. If I were to make an attempt at such a thing, I'd try to learn individual user preferences for content by observing their behaviour.
I think the review panel idea is another thing that shouldn't be at the level of the blockchain but instead the tools and ecosystem that surround the blockchain. To some extend what you're proposing (if I have it correct) is an up-powered version of curation.
If what you're proposing is at the level of the blockchain then bots can outscale the human reviewers and just accept the random loses. The economic inflection point for that is when the profit before a bot is discovered and black-listed is less than the cost of creating an account. I'm also not convinced the slowness of human review really work in a social network and there's the corruptibility of the reviewers to consider.
I think putting more tools and SP behind our human curators is probably going to acheive the same aims with less of the downsides.
Hey, great, any contribution towards improvements are welcome. There's also some tickets there ... juz sayin' 😜
I would be ideal if the filtering were done as a UI task, though obviously not client side. With the reorganization of the Steem services with Hive and so on, I think it will be offloaded from the basic blockchain operations, which are far too tightly coupled with it's internal node database and the RPC services on that. I think we need to wait for that but there's a chance to contribute directly to that code and perhaps really make what we're talking about here happen. I'd wait for a beta release though before I start poking around so as not to waste my time!
I had the opportunity to put an idea related to the review panel to Ned recently and he believes SMTs will take care of this. After considering it I'm inclined to agree, either that or it's an idea for something else, not Steemit after all.
On the idea itself though, there are many holes and I'm constantly reworking it. The latest best version is the creation of a curators SMT which uses a network of willing reviewers to kind of super charge a post. So instead of changing how Steem fundamentally works, it's an additional layer of reward, avouchment and community building. Speaking of which, it will be interesting to see how communities affect things.
On the AI accepting random losses, you can work around it I think. For example a multiple choice out of 4 with one right answer you have 25% change of randomly selecting the right one. But if you negatively score accounts that get a loss you can reduce the 1/4 takings to almost nothing if they choose uniformly randomly. Of course a human cabal of answers could collude against the system so I'm trying to come up with a way to build a chain of responsibility between reviewers as a way to mitigate this and the bot attack.
I'm coming up with a beta version (or a general system for testing) that might be able to operate in a live simulation mode without SMTs as they're not available yet.
I think your idea sounds promising. But, yeah so much wait-and-see: SMTs Hive.
A reputation based system for reviewers - sort what curie has, but not so monolithic might work.
It's a pity but yes, we must wait and see. The balance there is that any implementation would take at least as long as it will take to see some of these changes, and would perhaps become obsolete or in need of so much update that it'd be better to wait.
However I do think we could prototype some of these ideas before that time, perhaps using a BitShares asset or just a simple DB or on chain version of that by adding comments to a particular post, etc. It wouldn't have real assets but the ideas could be tested.
There's also the SMT test net, though it looks complicated and unfinished.