The dTAO mechanism is out of control? Bittensor is deviating from AI

Original title:Why dTAO is broken
Original author: @tzedonn, crypto writer
Original translation: zhouzhou, BlockBeats

Editor's note: Bittensor's dTAO mechanism was originally intended to distribute TAO releases more fairly, but the problem was exposed after only one month of its launch. The SN28 subnet exploits mechanism loopholes, pushing TAO to release into meme currency speculation, and is eventually centralized by the foundation. As decentralization advances, similar events may not be prevented. Bittensor may become a general incentive network rather than an AI project. The core problem is the lack of unified goals.

The following is the original content (to facilitate reading comprehension, the original content has been compiled):

I'm a person who has no resistance to novel token economics, and watching crypto protocols constantly adjust the incentives, sometimes seem clever -until they inevitably go wrong, the process itself is interesting. So when Bittensor launched the dynamic $TAO (dTAO) system on Valentine's Day (is this a Valentine's Day gift from @const_reborn?), I was immediately attracted.

The idea is simple: provide a new and more "fair" distribution method for TAO releases, so that each subnet can obtain TAO more reasonably.

But less than a month later, the problem emerged. It turns out that a design that looks reasonable does not necessarily survive the free market.

dTAO operation mechanism

A brief review of how dTAO works:

1. Each subnet has its own subnet token ($SN), which exists in a UniV2 type native liquidity pool (LP) based on TAO-SN. However, the "staking" TAO here is actually the same as "sending" TAO to SN. The only difference is that you cannot add additional liquidity to the liquidity pool, nor can you directly trade between different SN tokens (such as SN1 → SN2), but you can indirectly exchange it through TAO (SN1 → TAO → SN2).

2. The TAO release amount is allocated according to the SN token price of each subnet. They use moving average prices to smooth price fluctuations and prevent artificial manipulation of prices.

3. The release of SN tokens is also very high, with a supply cap of 21 million, the same as TAO and BTC. Some of them enter the TAO-SN transaction pool, and the rest are allocated to the stakeholders of the subnet (miners, validators, subnet owners).

4. The number of SN tokens entering the TAO-SN trading pool depends on how many SNs are needed to "balance" the TAO entering the pool, ensuring that the price of SN under TAO denomination remains stable, while liquidity increases.

5. However, if the number of SN obtained by the subnet exceeds its maximum releaseable amount (depending on the SN release curve), the SN release will be limited by the upper limit, causing the price of SN to rise under TAO pricing.

Core assumptions of the dTAO mechanism

Point (2) in the dTAO mechanism is based on a core assumption that a subnet with a higher market capitalization creates more value for the Bittensor network and therefore should obtain more TAO releases.

But the reality is that in the crypto market, the most expensive tokens are often those that are most popular, hyped, the most marketing, and even Ponzi-related. This is why the valuation of L1 public chain and meme coins is always relatively highest.

The original intention of mechanism design is good because it assumes that the subnet that truly creates value will repurchase SN tokens by generating revenue, pushing up the SN price, thereby gaining more TAO releases. But I think this logic is a bit naive.

meme coin subnet & collapsed token economy

Before dTAO went live, I discussed with several crypto analysts the obvious loopholes in the dTAO token economic model - high market value ≠ high income, which does not mean that more value is really created.

But I didn't expect that this theory would be verified by the market so quickly. The way free markets work is always unexpected.

Just before the dTAO upgrade, an anonymous user took over subnet 281 and directly turned it into a meme coin subnet, named "TAO Accumulation Corporation", referred to as "LOL-subnet". This obviously has nothing to do with AI.

On the (now deleted) Github page, it once wrote...no mining, just hold coins -completely turned into a capital disk

In LOL-subnet (subnet 281), miners do not need to run any code, and the validator's scoring mechanism is entirely based on the number of subnet tokens held by the miners. The more SN28 tokens you hold, the more TAO you get.

What actually happens is: speculators buy SN28 tokens → SN28 price rises, SN28 price rises → Get more TAO releases, if the TAO release exceeds the token release limit of the subnet → SN28 price continues to rise, the released SN28 tokens are distributed to "miners" according to the holding ratio → The more SN28 holds, the more you get. In order to obtain more TAO, more people buy SN28 → The higher the price → Ponzi cycle continues.

In the end, the TAO release amount officially flowed to...meme coins! At one point, SN28 even became the 7th subnet of Bittensor's market value.

Why didn't SN28 take over Bittensor completely? Centralized action saved Bittensor.

Within a few days of SN28's rapid expansion, the Opentensor Foundation directly used their root stake to run custom validator codes, encouraging everyone to sell SN28. In the end, SN28 plummeted 98% in a few hours and was completely liquidated.

SN28 plummeted 98% after the Opentensor Foundation took action.

Essentially, the Foundation acts as a centralized entity, preventing the free market from operating on its own in accordance with the dTAO mechanism. But the reason they can do this is because they are still in a transition period -the release mechanism of TAO is gradually changing from the old model to the dTAO mechanism.

Old TAO Release Mechanism & Transition to DTAO

Under the old mechanism, 64 validators holding the most TAO pledges in SN0 ("root subnet") can vote to determine where the TAO release will go.

But this mechanism also has huge profit-driven problems, especially when big validators (such as Opentensor Foundation, DCG Yuma, Dao5, Polychain, etc.) have too much power.

For example, potential conflicts of interest include:

They can prioritize allocating TAO to subnets they invest or incubate.

We can direct the TAO release to the subnet where we run the validator and can get the TAO reward.

These problems have existed long ago, and dTAO originally wanted to solve this centralized problem, but the SN28 incident proved that the new mechanism still has major flaws.

Going towards the right direction of decentralization, but the risks are still there

Getting rid of the old mechanism is indeed the right step towards decentralization. Although this may mean that the team will lose some control over the TAO release, I still agree that they chose a more decentralized reward mechanism.

However, when the SN28 incident occurred, the dTAO mechanism had just been launched for a week, and SN0 (root subnet) still controlled about 95% of the TAO release (see the blue line in the figure below), which allowed the Opentensor Foundation to intervene quickly and prevent the continued expansion of the capital market.

But the problem is:

·After about a year, the power of SN0 will gradually drop to around 20%, by which time it will no longer directly control most of the TAO releases.

·If there is another SN28 situation in the future, it is very likely that no one will be able to intervene through SN0.

In this case, Bittensor may no longer be a "decentralized AI" project, but will completely become a meme coin incentive network.

Bittensor is still in the transition period of the release mechanism, and control is gradually shifting from the old mechanism (SN0 or "root prop") to the new mechanism (dTAO or "alpha prop").

Not just meme coins, Bittensor may become a general incentive network

Even if we assume that in a bear market environment, people will not rush into meme coins, Bittensor still has a high probability of becoming a "pan-incentive network" that has nothing to do with AI.

For example: If someone launches a subnet for decentralized Bitcoin mining (the concept is not new), it can obtain the TAO release by incentivizing more efficient BTC mining methods and then using the mined BTC as continuous revenue to repurchase SN tokens.

If this model is established, TAO will change from a decentralized AI project to a general incentive project, and the release of TAO is no longer used to promote AI development, but has become a subsidy means for various commercial operating costs (OpEx).

Technically speaking, this is not wrong in itself, because the Yuma consensus mechanism is originally intended to reach a consensus on "subjective" work and is not necessarily limited to AI. But without a clear goal, the entire network could become… meaningless.

Final Chapter: The cracks in the dTAO mechanism have appeared

The dTAO mechanism has only been online for one month, and the problem has emerged.

According to the incentive logic of the free market, without centralized forces intervention, Bittensor may no longer be an AI project, but an "attention network", dominated by the meme coin subnet, or become a general incentive network, where various companies use TAO to release subsidized operating costs without promoting the development of the Bittensor ecosystem.

I think Bittensor needs a real "objective function" to get all subnets to move in the same direction. But the problem is that it is difficult to define an absolute goal (AGI?), as we have seen, even the LLM evaluation framework is difficult to achieve complete fairness... This is why the Yuma consensus mechanism was originally intended to form a consensus on "subjectivity" work.

"Tell me the incentive mechanism and I can tell you the result." Peace!