FUD is rampant, will the new AI king Bittensor fall from grace?

Author: Deep Tide TechFlow

One generation's narrative, one generation's deity.

Since the end of last year, with the rise of AI narratives, Bittensor has emerged, and its token TAO skyrocketed from around $80 in October last year to a peak of $730 in March this year, with a market cap reaching about $4.7 billion.

Creating gods in crypto is terrifying.

Although it hasn't surged as absurdly as meme coins, Bittensor has indeed become the new king in the AI sector over the past six months, even ranking among the top 30 cryptocurrencies by market cap.

From being ignored to being too afraid to invest, and then to a surge that caught everyone off guard on Binance, the market has indeed heavily invested in this project.

But apart from Bitcoin and Ethereum, no crypto reign is eternal.

From the price peak to now, the TAO price has fallen back to around $450, and more and more FUD about TAO has gradually emerged in the market; various doubts and criticisms seem to be shaking the foundation of this new AI king.

As the viewpoint that "all AI coins are memes" becomes increasingly popular, market participants have woken up and are scrutinizing this project at the peak of its sector with a magnifying glass.

Keen hunters have also sensed opportunities, trying to exploit every loophole to profit from the demystification campaign aimed at bringing down the new king.

We have gathered the FUD voices surrounding Bittensor in the market to see if there are any hidden issues beneath its glamorous exterior.

Resource Waste, Doing Useless Work

In our article "Interpreting Bittensor (TAO): An Ambitious AI Lego that Makes Algorithms Composable," we mentioned that Bittensor does not produce algorithms; it merely acts as a transporter of high-quality algorithms, filtering out the best AI algorithms or models through a market incentive mechanism.

In the open AI supply-demand chain of Bittensor, some provide different models, some evaluate different models, and some use the results produced by the best models.

However, reality seems to be more disappointing than ideals, as resource waste and doing useless work are quite common in the process of selecting and evaluating AI models.

Twitter user @ercwl pointed out that subnet 1 of TAO is engaged in filtering "the answers corresponding to the best text prompts." Users or systems generate a query or question (i.e., text prompt), and each node (miner) in Bittensor is responsible for running one or more machine learning models (such as large language models, LLMs) to generate answers to the text prompts.

Clearly, whoever provides the better answer gets used; however, the situation is worse than imagined.

Miner nodes in the network run their models on these text prompts to generate answers. Each miner may use different models or model configurations to respond to the same prompt.

To ensure the accuracy of the answers and the decentralized nature of the network, multiple miners independently generate answers. This leads to the first redundancy of resource waste: the same question is processed multiple times.

Additionally, validator nodes are responsible for evaluating the answers provided by miners. Since each miner tries to generate answers that are closest to the validators' expectations, they tend to adopt similar strategies and model configurations. This means multiple nodes are performing nearly identical computational tasks, which are actually redundant, leading to resource waste.

For example, this subnet is filled with basic questions like "What is water?" which may be answered simultaneously by hundreds or thousands of miners with an obviously common-sense answer, such as "Water is a compound with the chemical formula H2O."

Since the system verifies and rewards miners by comparing the similarity of answers, this results in significant redundancy, as multiple miners are merely repeating work that others have already done.

Using decentralization, the most resource-intensive way to reach consensus (because getting a group of strangers to agree requires more resources), to validate a large number of similar common-sense Q&A, in order to gain incentives, leads to answering just for the sake of answering, wasting the computing power provided by OpenAI or other LLM providers, as well as the costs paid for APIs.

Not as Decentralized as You Think

Another AI project, Hyperspace, CEO @varun_mathur pointed out that Bittensor may not be as decentralized as many think.

In many cryptocurrency networks, preventing attacks typically requires control of more than half of the network (51% attack). In the case of Bittensor, the network may allow a lower control threshold (40%), significantly lowering the barrier for conducting an attack and increasing the risk of the network being manipulated or controlled by a few large nodes.

Varun believes that Bittensor's degree of decentralization is insufficient, as ownership of 40% of TAO tokens (which can be held by just three entities) could launch an attack.

If the three major nodes of Bittensor control more than 40% of the validation power, these nodes may conspire for their own interests, validating each other's transactions or data to gain improper benefits. This behavior not only undermines the fairness and transparency of the network but could also lead to security incidents, as it provides opportunities for potential malicious actions or attacks.

Moreover, another issue of insufficient decentralization is also reflected in the "weight setting" of validators towards miners.

In the Bittensor network, weight typically refers to a numerical value given by validators to miners, based on the consistency of the answers generated by miners with preset standards or reference answers. The higher the weight, the more the miner's answer aligns with the validator's expectations, and correspondingly, the more TAO tokens the miner receives as a reward.

Since weights are set manually, this can introduce subjectivity and manipulation issues. If validators favor certain miners or if the weight setting is not transparent and fair enough, the overall trust and efficiency of the network may be affected.

Clearly, under this design, small players have no room for survival:

Small or independent miners may struggle to achieve high weights due to a lack of sufficient resources or technical support, which could lead to further centralization of the market, making it so that only large or well-funded miners can receive adequate rewards.

Dragon-slaying warriors ultimately become dragons themselves, defeating centralized AI but unable to eliminate centralized "human governance."

Token Centralization, Selling Pressure is Tense

StinkyInsect Labs leader Comrade Xiao Wang @0xInv1ctus directly pointed out in a post titled "My Big Poster - Why I Must Criticize BITTENSOR" that the allocation of TAO tokens is shrouded in doubts of being highly controlled by internal small groups, with risks of sell-offs at any time.

The basis of this FUD lies in the unsettling nature of TAO tokens from their inception to their results:

The $TAO token began to be produced in 2021, but there is no information explaining how the tokens produced from January 3, 2021, to October 2, 2023, when the subnet went live, were allocated and where they ultimately flowed.

From the results, the staking amount of the top 12 root network validators accounts for 79% of the entire network. Moreover, based on the public business statements of these individuals, they only engage in staking activities related to Bittensor, raising reasonable suspicion that these validators, holding over $20B worth of $TAO, maintain close relationships, forming an internal small group.

What is most unsettling is that Bittensor's staking has no lock-up period, allowing for immediate withdrawal, meaning that the currently staked tokens, which account for 85% of the total circulating supply, can be sold at any time.

At the same time, the hedge fund of the author's institution has publicly stated that it is shorting TAO at around $420, expecting its price to return to $100-150.

One generation's narrative, one generation's deity; no reign is eternal.

As the new king created by the AI boom, Bittensor will attract attention and liquidity, but it is also destined to see them flow out.

Perhaps, as these FUD statements suggest, the business and practical implementation corresponding to TAO simply cannot support such a high market cap and price; but when the outflow occurs, whether you can profit from shorting during the exit, there is actually no standard answer.

Maybe just like the classic horse-drawing MEME below, even if you know full well that there is a world of difference between "what the project wants to do" and "what the project can do," you still cannot make a fortune during the narrative rise with more information, more seasoned trading instincts, and more steadfast trading discipline, and swim away unscathed when the tide recedes.

Who is really making money in this market?

Crypto retail investors are becoming increasingly confused amidst waves of FUD statements and posts boasting "I am financially free."