Programming Diary #37: Harnessing AI for lifetime rewards and to enhance the Steem ecosystem

in Steem Dev3 days ago (edited)

Summary


@elonmusk on X (Twitter)

In today's post, I describe recent progress with the Steem Curation Extension and Thoth projects. The Steem Curation Extension has been switched to beta and published in the Chrome web store. Thoth has been updated with screening improvements and with the ability to start directing beneficiary rewards to delegators.

By including authors and delegators in the beneficiary pool, the hope for Thoth is that it will harmonize the interests of creators and investors and reduce or eliminate the dysfunctional incentives that are created by gen-4 voting services. These dysfunctional incentives currently place investors and authors at cross-purposes, and I believe they're driving the creators out of the ecosystem. Additionally, Thoth will provide fully passive rewards to investors, eliminating the need for investors to post vacuous content daily just so they can harvest their gains.

In addition to describing activity on my existing projects, the post also contains reflections on some recent influencer statements on AI technologies and how those statements bear on the Steem ecosystem. In short, I believe that rapid adoption of AI is a critical and urgent need for the Steem ecosystem.

Background

Here's what I'll be thinking about for today's Reflections section.

1.if you are not skillsmaxxing with o3 at minimum 3 hours every day, ngmiSam Altman, OpenAI CEO, on 𝕏.
2.The 𝕏 recommendation algorithm is being replaced with a lightweight version of @Grok, so will soon be dramatically better!Elon Musk on 𝕏.
3.AI will become a million times more powerful during the Trump presidency. People find this hard to believe. Let’s break it down.David Sacks, White House A.I. & Crypto Czar on 𝕏
4.Some groundbreaking AI will be developed on Tron and @steemit. Stay tuned.@justinsunsteemit on 𝕏, January 9, 2025

Activity Descriptions

For this week, I have updates on two of the tools that I have worked on. The Steem Curation Extension (Chrome Store, github), and Thoth.

Steem Curation Extension

The good news about the Steem Curation Extension is that it's now available in the Chrome store, and the latest version includes some nice new features from @cmp2020. The bad news is that zero people have downloaded it in the week and a half that it's been available (besides me, testing to make sure that it works).

Personally, after using the extension for a number of years, visiting a condenser web site feels incomplete without it. I now expect to be able to find my own voting power in the profile icon, enable and disable resteems, and see enhanced metrics about the posts and authors in my feed, so I hope that others will also try it out and find it useful.

For this tool, the big activity during the interval was preparing it for review by the Chrome store and trying to start some sort of a more disciplined release methodology. To that end, I worked with Google Gemini to create two checklists for activities with each new release. The versioning checklist is intended to get me through the process of publishing a new release in github, and the pre-submission checklist is intended to check all the things that might block approval from the Chrome webstore. In all, on a day off from my day job, I spent about 13 hours creating the checklists and running through them. Fortunately, this translated into approval from the Chrome web store after about 3 or 4 days of review. From web searches and LLM interactions, I had expected that the review process could take up to 28 days, so receiving approval in less than a week was a nice surprise.

Thoth

In my previous diary post, I said that I hoped to get started on directing rewards to delegators. Here's what I wrote:

I'm also thinking about doing a first pass at providing beneficiary rewards to delegators. If I let Thoth find 5 posts, that only leaves me one free beneficiary slot for delegators, so I'm thinking that I'll select the delegator randomly, with their odds based on the delegation share. This method is far less than ideal, but I think it's suitable and fair for a small number of delegators. And, it's the only option that I see right now. The 8-beneficiary soft-limit really makes this challenging.

I accomplished this. So basically, now, I can distribute rewards between authors and delegators, with a total of 6 between the two groups. (Two of eight slots are taken by @null and by the posting account.) For testing purposes, I have now settled on discovering 3 posts, in which case the tool can direct rewards for 1-3 randomly selected delegators - where the odds of inclusion are proportionate to their delegation amounts.

In addition to adding payouts to delegators, I also backed out the previous changes that had given the AI too much discretion, and I updated the follower network screening to incorporate a "half life" penalty for old accounts, where some portion of their followers have likely become inactive.

Next Up

As long as no problems emerge, I'm done with the Steem Curation Extension for a while.

During the coming interval, my attention will be focused on Thoth. Specifically, I will be focusing on activating and enhancing more of the pre-AI screening capabilities, and maybe also adding some other information to the post that Thoth produces and trying to improve the aesthetics.

Also, I want to move the AI prompts out of the code and into files of their own, so that Thoth operators can customize their AI prompts without requiring a new code release.

Reflections

AI in social media

Thinking about those quotes in the background section, what do they have to do with Steem? In short: Every day where AI doesn't improve its ability to advance the blockchain's interest in the Steem ecosystem is a day where the Steem blockchain is falling behind. And at the pace of AI growth, each lost day might be more expensive than the day before.

Personally, I think that the quotes from Altman and Sacks are probably overstating the near-term potential for AI, but the underlying point is still true. Even if LLM capabilities were frozen in place at today's level (not likely), there's still years or decades worth of improvements that could be built on top of them in our ecosystem, and hardware improvements are definitely coming. In Sacks' argument, he justified the million-fold increase on the convergence of three factors:

  1. Advances in models: 3-4X per year.
  2. Chip improvements: 3-4X per year.
  3. Data centers: 100X in 4 years

My guess is that Sacks is overstating the rate of advance in the models. But, even if he's wrong about the overall magnitude, he's not wrong about the direction.

To me, the current state of AI is like the Internet during the dotcom era. When the dotcom bubble burst, we were left with massive amounts of dark fiber that sat idle while people figured out how to use it. Eventually, though, people did figure out how to use it.

So, as a stakeholder in the Steem economy, I want to see us making prudent use of this new foundational tool. I don't want to see reckless, dotcom bubble level investment, but responsible growth is necessary. @justinsunsteemit said to "stay tuned" back in January. It's May, now. I'm not sure if he and @tronfoundation are going to deliver on his claim, but the rest of us should.

  1. The only way to fight abuse at scale is with automation at scale. Humans are not suited to doing it without the right tools.
  2. Roughly 99.99% of Steem's content is (mostly) ignored by the 7-day reward payout mechanism. That percentage grows daily. This means (roughly) that the rewards mechanism is focused on 0.01% of Steem's value to the attention economy. The task of deriving value for the ecosystem from that other 99.99% is not well suited for humans without the assistance of AI.
  3. As mentioned in last week's post, the Steem platform could encourage and reward vibe coding of tools for the ecosystem.
  4. Note how Thoth is trying to do a similar thing to what Elon suggested in his quote, "The 𝕏 recommendation algorithm is being replaced with a lightweight version of @Grok, so will soon be dramatically better!". Thoth wants to find interesting content, put it in your feed, and direct rewards to the people who made that possible.

If Steem's stakeholders want the value of our investment to go up, IMO, responsible use of AI is one of the most promising and urgent ways to make that happen. If the "bad guys" are using AI, and the "good guys" aren't, then the "good guys" don't stand a chance.

The math and incentives of Thoth

Beneficiaries, authors, and posts per day

One of the things that I mentioned above is that I have reduced the number of discovered articles per @thoth.test post from 5 to 3 during testing. One reason for that is the soft limit of 8 on the number of beneficiaries. By reducing the discovered authors from 5 to 3, I increase the available beneficiary slots for delegators from 1 to 3. There are other reasons, too, though.

One reason is that some undesirable articles do find their way through. This includes plagiarism, AI generated content, and just plain bad writing. When this happens, it can discourage votes and pollutes the entire post. This harms the other authors who are included in the same post. So, lowering the number of posts is a way of limiting the damage when that happens.

The main reason is this, though. If something like Thoth is going to have any chance of outcompeting the gen-4 voting services, it needs to operate at scale - this means 10 posts per day. The difference between discovering 3 and 5 articles per post is the difference between 30 and 50 discovered articles per day. This number of articles might have been supportable during the early years - say, 2016 to 2018, but in recent years, I just don't think we have enough attractive content to support that rate of discovery.

With the same screening in place, Thoth finds 3 articles in a few hours during 2016, but it can take half-a-day in 2025.

It can be hoped that lifetime rewards payouts will improve the quality of content that appears on the chains, and the number of articles per post could eventually be increased again.

Active or historical

Thoth is able to review currently active articles or historical articles. So, the question is how much of each should be covered. The main benefit of supplemental rewards to active articles that I see is that the authors are still here, so we know that the rewards are going to active contributors. The main drawback is that we don't really seem to have enough active authors who are producing articles that will deliver clicks from the search engines. Also, all rewards in the ecosystem are already directed towards active authors, so it makes sense to shift some portion to the area where the hidden value lies.

On the other hand, when we direct rewards to historical articles, they are quite frequently going to inactive authors. If those authors never come back, this is equivalent to burning those rewards. For example, I'm aware of at least one deceased author to whom @thoth.test has directed rewards. Benefits, however include (1.) new life for articles that would have remained hidden; and (2.) The potential to reactivate authors who had left the platform.

As noted, the two possibilities aren't mutually exclusive, because active posts become historical posts after 7 days.

So, during testing, my current balance is to use a random start date 6 days per week and an active start date once. This may change.

Conclusion

In this article, I have provided updates on the two tools that received my attention in the last couple weeks - the Steem Curation Extension and Thoth.

The Steem Curation Extension has now been published in the Chrome web store, making it more accessible to the Steem community. I hope that Steemizens will find the enhanced information and other features to be valuable additions to the Steem experience.

Regarding Thoth, I made some progress with the ability to add delegators to the reward distribution. Hopefully, this will be a step towards the alignment of incentives for content creators and investors within the Steem ecosystem. I also implemented some refinements to Thoth's content screening in an effort to reduce the impact of undesirable content like plagiarism and low-quality posts.

Looking ahead, I believe there is a critical and urgent need for responsible integration of AI technologies that will advance the Steem blockchain's interests. Some ways that AI can be brought to bear include the discovery and reduction of undesirable content; directing supplemental rewards to valuable and attractive content; and rewarding vibe coding in order to bring new tools to the ecosystem.

*At some point, I eventually get tired of proof-reading. That's the time to hit "Publish"😉.

Community Follow-up

  1. Please install the Steem Curation Extension and let me know your thoughts about the enhanced information that it gives you.
  2. Please check the posts from @thoth.test and let me know your thoughts on those? Is it catching any that look plagiarized or AI generated? Even though it's still in testing, you can already send rewards to the authors it finds by upvoting the posts.

Thank you for your time and attention.

As a general rule, I up-vote comments that demonstrate "proof of reading".




Steve Palmer is an IT professional with three decades of professional experience in data communications and information systems. He holds a bachelor's degree in mathematics, a master's degree in computer science, and a master's degree in information systems and technology management. He has been awarded 3 US patents.


image.png

Pixabay license, source

Reminder


Visit the /promoted page and #burnsteem25 to support the inflation-fighters who are helping to enable decentralized regulation of Steem token supply growth.

Sort:  

I downloaded your extension yesterday and was pleasantly surprised by how informative it is. I think people just don't understand how useful it is and think that it is only needed for curation teams, which is not true at all.

 2 days ago 

Thanks for the feedback. Please let me know if you run into any problems or have any suggestions. I can't promise that updates will be fast, but I'll definitely do my best to incorporate fixes and suggested improvements.

Hey bro, this is the first time I've seen 'Promotion Cost' . What is it?"

image.png

 2 days ago 

The blockchain has a built in promotion mechanism. Before payout time, if someone sends any amount of SBD to @null with a memo in this format:

Then the post will show up on the /promoted page. The list is sorted by total promotion amount (in descending order). Each community and tag has its own sub-page with promoted posts, too. i.e. Here is the "promoted" page for Steem-Dev.

More on that, here and here. This and @null beneficiary rewards was the origin for the Steem Curation Extension. It was introduced at about the same time as the launch of the #burnsteem25 program by SC01.

By default the only way to see that a post has been promoted is to go to that page, and nobody ever remembers to go there. So, I built a browser extension that tells me at a glance if anyone is promoting posts in my feed or making use of @null beneficiaries, and I make sure to give those posts curation attention. The turquoise outline that you see in the post above is added to promoted posts by the browser extension. It doesn't exist in condenser.

interested. There are many things that are hard for new users to notice because they're outdated.

 1 hour ago 

Agreed. It's possible to learn about much of it by reading the historical posts from @steemitblog, but many of those posts have probably also become outdated.

It is indeed interesting, thank you for explaining.

Thank you for asking.

 2 days ago 

Thank You!