As somebody who curates manually, and also likes your idea, what would the best way of supporting this be? Presumably in my case, I'm best to keep my STEEM Power, rather than delegate it to you? Although without a decent SP base, can thoth have a big enough impact?

The original thought was for active posts. In that case, I imagined that you'd be able to scan the output for posts that look interesting and then click through and support the featured authors directly. Now that historical posts are in the mix, it gets more complicated. Of course, you could vote the Thoth post - but I get that there may be disqualifying content in some. You could also use an eversteem style comment/beneficiary - mechanism to individually support any interesting paid-out posts that it highlights (maybe I could even consider adding one reply to the Thoth post with an individual beneficiary for each included author? Not sure if I like that idea, though...).

For historical posts, if I find a way to deal with the cap on beneficiaries, then hopefully Thoth can get enough delegated for passive income in order to make a difference. Besides ROI and user retention, I suppose there are may also be SEO benefits from resurfacing and relinking lost content(?), so passive investors have multiple self-interested reasons to delegate.

Given that there are already people highlighting "active" content, should thoth focus on content that isn't active and has already paid out?

This is a good point. I think that's probably the case. At least near the beginning. It's funny how the project seems to be getting flipped upside down.

In my opinion, the posts selected don't highlight the best of Steem in the past or at present so I think the selection algorithm will need some adjustment. Proof of Concept though and in my opinion, it's looking good.

I completely agree. In addition to improving the post selection, I also need to stop it from picking the same post multiple times when someone edits an existing comment. The current version is definitely proof of concept.

I also wonder if Thoth could self-learn by analysing the success of each post it produces?

Definitely possible in the future. @cmp2020 had a machine-learning voting bot running a couple years ago when he was in school. Eventually, that sort of selection could be inserted between the screening and the LLM - or before the screening. That's a long way down the road, though, I think.

I'd also think about attention span. I'm not lazy but I'm "Time Poor". So it's too text heavy for my liking.

I agree on this point, too. I should have mentioned that in the next steps section. I'm going to have a table near the front with just the titles, authors, dates, tags, and maybe 100 words from the body, then I'll put the AI output in a separate section - maybe another table. That way the reader can scan the summary table first and decide which of the AI summaries they want to read.

You could even introduce some bias now based upon your own opinions of the /tags page.

Include/exclude of tags is one of the screens that I'm still planning to add. I think bias is good. I'd like to see multiple instances of Thoth all running with their own unique biases and then the community can decide which ones are most valuable.

Final thought - It's a shame that @thoth went to some pointless account 9 years ago.

Yep. That was one of the first things I checked, too. Disappointing. I have an alternate in mind that is available, but @thoth would've definitely been ideal. Thank you for the feedback!