Preliminary observations about Thoth economics

Q1: How much does it cost to run the LLMs that power Thoth?
A1: Someone has to provide the computer for Thoth along with the manpower to run/schedule it, but aside from that, the cost is zero: $0.

Free APIs from Google Gemini or ArliAI can be used.

As you can see, the usage rate fits comfortably into the Gemini free tier, even for Gemini-2.5-Pro.

(The usage shown covers about 2 or 3 runs per day - which is enough to optimize voting power, plus some additional testing.)

Q2: What sort of return does it give to delegators?
A2: It depends, and it will always change. We have some real-world numbers to look at now, though. For comparatively large delegators, we can use Steemworld to compare pending curation rewards (from auto-voting) and beneficiary rewards from penny4thoughts.

Curation:

Beneficiaries:

So, 9 STEEM/SP for delegation vs. 26 SP for curation. And penny4thoughts has a little more than 2x dedicated to curation vs. delegation. So, at present that account's delegation is generating about ²/₃ as many rewards as automated voting.

For comparatively small delegators, we can look at my wife's account, lisa.palmer.

Curation:

Beneficiaries:

That account has roughly the same amount of SP dedicated to each purpose: auto-voting and delegation. It is set to receive 0.7 SP for curation vs. 3.2 STEEM/SP for delegation. So, delegation is presently outperforming curation by a long-shot for small delegations.

I could be wrong, but my expectation is that if the number of delegators were larger, it would tend to balance out, so that performance would be similar for all accounts. This is because delegators get placed into posts as beneficiaries with a likelihood that's proportionate to their delegation amounts. I believe the discrepancy in the test account is likely due to the small number of delegators, which means everyone is included in all replies. This favors smaller delegators, tipping the scales in their favor.

---

Q3: Why did the test account switch to posting with 50/50 instead of 100% SP.
A3: Because some of the rewards are being sent to accounts that are likely dead - lost keys, etc... Sending SP to those accounts will reduce interest rates that go to active accounts, so I decided to minimize the SP distributions. (of course, any Thoth operator could set that to their own preferences)

---

Final note: Over the weekend, I read the article, There Is No A.I., by Jaron Lanier, which raised a fascinating concept. I'm sure I've written about it here, before, but this article refreshed my memory on Lanier's data dignity concept.

The basic idea is that AI is built on a foundation of human creativity, so the automaton should remember and compensate the sources of information that it makes use of.

If we buy into this argument, then maybe there's a moral case to be made that some form of #lifetime-rewards is a necessary enhancement for the Steem blockchain.

Here's an embedded YouTube video with Lanier discussing the concept.