Editor's note: The author introduces the Bittensor ecosystem, a Web3 platform that promotes the development of decentralized AI through "Darwinian AI". The author shares his passion for cryptocurrency and AI, and discusses the convenience and potential risks of centralized AI products, such as data ownership and platform stability issues. Bittensor promotes the natural selection and evolution of AI models through competition and incentive mechanisms, using $TAO tokens and subnet structures, attracting investors' attention.

The following is the original content (for easier reading and understanding, the original content has been reorganized):

Cryptocurrency has always been fascinating to me. There is always something new to learn. I am naturally curious and love asking a lot of silly questions to technical people just to get their insights and learn from their valuable experiences.

Artificial Intelligence (AI) is no exception, in fact, things are moving incredibly fast. Web2 tech giants are constantly improving their models, and major applications are leveraging AI to launch various AI-driven use cases:

@canva launched AI tools that allow non-technical artists and creators to easily build interactive experiences and enhance their creations with AI.

· @YouTube has introduced a new AI tool that allows creators to generate background music for their videos.

Ride-hailing platforms like Grab have deployed agentic AI to support merchants and driver-partners.

E-commerce platforms like Lazada have introduced generative AI (GenAI) tools to help sellers improve sales, marketing, and customer service.

The list goes on. Real-world use cases for leveraging generative and agent AI to improve workflows are continuing to gain adoption among enterprise and retail users.

The beauty of these technologies is that they are readily available—you can find free or low-cost solutions almost everywhere. The benefits far outweigh the financial costs.

But people often overlook the hidden trade-offs when using these AI products, such as:

Who owns your data?

Could someone else take your idea and create a competing product?

Is the platform secure? Will your data be leaked?

If the platform goes down (like AWS did), will your business be shut down? Will customer funds be at risk?

Can you always access your platform? Do you need to verify your identity? If the platform shuts down, can you still own your product or business?

There are many more issues (I discussed these in more detail in my previous post if you haven’t read it).

Centralized players have centralized power, and their decisions can (unintentionally) have a huge impact on your life.

You might say that doesn’t matter — maybe you don’t use these tools often, or you trust these companies to act in the best interest of their users. That’s fine. You might even want to invest in these AI startups because they’re tapping into huge markets. But here’s the thing — you can’t. Unless you’re at @ycombinator or a top VC firm, you don’t have access to these investment opportunities.

On the other hand, in Web3 AI, there are many investable AI ecosystems, and the team is committed to bringing decentralized AI products and services to users. One of the most worthy decentralized AI (DeAI) ecosystems is @opentensor (Bittensor).

Bittensor: Darwinian AI

Bittensor falls into the category of "Darwinian AI" - driving the evolution of AI through natural selection. Imagine this is an AI version of "The Hunger Games", where each subnet has its own "Hunger Games" and "miners" are contestants (or "tributes"). They compete with each other on specific tasks with their own models and data. Only the most adapted models (the ones with the best performance) will be rewarded. Weaker models will be replaced or evolved (through training, adjustment, or learning from other models). Over time, this will form a more powerful, diverse, and high-performance AI ecosystem.

What’s particularly exciting about Bittensor is its competition and incentives mechanism, which aims to align incentives between different stakeholders. I outlined the challenges facing the Web3 AI Agent team in the tweet below…

tl;dr: Current proxy tokens are a great tool for speculators and teams to create hype, but are bad for user acquisition and retention, and cannot serve as an incentive mechanism to retain talent (developers, founders, etc.), especially when prices fall.

Bittensor solves this problem with a market-driven mechanism that allocates $TAO emissions to subnets, thereby incentivizing and supporting the operations of teams. The market determines which subnets get more emissions by staking $TAO within the subnet. Once staked, $TAO is converted to Alpha subnet tokens. The more people stake, the higher the Alpha token price, and the more emissions you get (in the form of Alpha tokens).

TAO's emission mechanism is very similar to BTC, with a fixed total of 21 million tokens, which is halved every 4 years (7,200 $TAO are emitted to the subnet every day). The first $TAO halving is expected to occur around January 5, 2026, when the circulating supply will reach 10.5 million tokens.

Why this matters to investors

I won’t go into the technical details here — I just want to share why I think Bittensor is one of the most exciting ecosystems to look at from a trading/investing perspective.

In addition to the above dynamics, when trading Alpha Subnet tokens, it feels like trading and mining at the same time.

This is because every time the Alpha token price appreciates, you not only experience the price appreciation, but also receive $TAO emissions (in the form of Alpha tokens).

If a subnet performs extremely well and rises in the rankings, your initially staked $TAO will experience a massive price appreciation and emission surge. The earlier you stake $TAO to a subnet, the higher your annualized yield (APY) will be (because the market has not caught up yet, there are fewer people staking and fewer $TAO).

dTAO vs Solidly

Solidly's ve(3,3) requires long-term lock-up and continuous participation. Emission losses caused by wrong voting (voting for the wrong LP pool) are borne by all holders (emission is sold, and the price drops for all token holders).

While dTAO does not require long-term lockups, anyone can enter or exit at any time, staking on a subnet requires a lot of due diligence (DYOR). Investing in the wrong subnet can lead to huge losses (because people can easily exit and there is no lockup period).

However, the FDV (fully diluted valuation) is too high! How to invest in a subnet with a FDV of more than $500 million?

FDV might not be the best indicator here as the subnet is still in its early stages and market capitalization (MC) might be more suitable (if you are trading in the short to medium term).

If you’re concerned about inflation, it’s helpful to understand the 18%/41%/41% emission split - these are the emissions (in Alpha tokens) received by subnet owners, validators, and miners respectively. As a staker/Alpha token holder, you profit from the validator’s 41% portion, since you delegate $TAO to them when you stake.

Many subnet owners continue to hold the Alpha tokens they receive from emissions as a sign of confidence, and many are in active dialogue with validators and miners to keep them bullish on the project and avoid selling large amounts of tokens (this information can be viewed on taostats).

Zooming out, one of the best charts of trends within the Bittensor ecosystem is below:

Since the launch of dTAO in February, %TAO in Root (the original subnet that manages the Bittensor incentive system) has continued to decline, while %TAO in subnets has continued to rise. This means that stakers/investors are increasingly willing to take on risk (conservative APY for staking on the Root network is around 20-25% with no price appreciation of Alpha subnet tokens).

This trend is consistent with the speed at which subnet teams launch products. Since the launch of dTAO, the team needs to build publicly, develop products that users want, iterate quickly and find product-market fit (PMF), acquire users and quickly generate real utility and significant revenue. Since I entered this ecosystem, I can feel that the team's development speed is much faster than other ecosystems (due to competition and incentive allocation mechanisms).

This brings us to subnets and their unique, investable DeAI use cases.

Leading Subnets and Use Cases

The team considered best at shipping products with PMF, targeting regular people, and performing professional and continuous public builds is @rayon_labs

——SN64 (Chutes), SN56 (Gradients), SN19 (Nineteen).

Chutes — Provides infrastructure that makes it easy to deploy AI in a serverless manner. The best example is the recent AWS outage, which can bring your AI application to a halt (potentially leading to loss of funds or vulnerabilities) if you rely on a centralized provider because there is a single point of failure.

Gradients — Anyone with no coding knowledge can train their own AI models on Gradients (for specific use cases, image generation, custom LLMs). The recently launched v3 is cheaper than its peers.

Nineteen — Provides a fast, scalable, decentralized AI inference platform (available to anyone for text and image generation use cases, much faster than its peers).

In addition to this, Rayon is launching the Squad AI Agent Platform, an easy-to-use drag-and-drop node-style AI agent building platform that has generated widespread interest in the community.

Together these three subnets own over 1/3 of the total $TAO emission - a testament to the team’s ability to build openly and deliver quality products that users want (Rayon is hailed as the #1 team by many subnet owners).

Gradients grew 13x in one month (current market cap $32 million).

Chutes grew 2.3x ($63 million market cap).

· Nineteen grew 3x ($18 million market cap).

This trend doesn’t look like it’s going to stop anytime soon, especially with the adoption rate of Chutes (currently the #1 subnet).

In addition to the Rayon Labs subnet, there are many interesting teams - protein folding, deep fake/AI content detection, 3D models, trading strategies, role playing LLM. I haven't dug into everything yet, I think the easiest to understand are the subnets under "prediction systems" (taopill), especially:

SN41 @sportstensor

You may know them from @AskBillyBets. Sportstensor is the intelligence that powers Billy’s decisions (the core team leading Billy is @ContangoDigital, a VC that invested in DeAI and is also a validator and miner on the Bittensor subnet).

What makes SN41 unique is its product, the Sportstensor model. It is a competition between miners with the best models and datasets to predict sports game outcomes.

Example: In the latest NBA league, if you follow the crowd (favorites), you will have about 68% accuracy/win rate. Does this mean that everyone can make a lot of money on the favorites? No, they actually lose money. If you bet $100 on each favorite, you will end up with a negative return on investment (ROI) and a loss of about $1,700.

Although the favorites have a higher chance of winning, the better odds mean you make less money when you guess right. People tend to bet on the favorites, which leads to low odds for the underdogs, which means you can make a lot of money if you pick the underdogs right.

This is where the Sportstensor model comes into play. Miners run their own machine learning models (Monte Carlo, Random Forest, Linear Regression, etc.) using their own data (free or proprietary) to get the best results. Sportstensor then takes the average/median of these results and uses it as intelligence to identify advantages in the market.

The actual market odds may be 25:75, and the model may show odds of 45:55. This 15 difference is the advantage. If the model finds many such advantages, you can accumulate positive ROI in the long run without a high win rate.

Check out their full deal report (if you want to dig deeper):

This is the model result shared in their latest report, and the data is quite amazing. The team also runs a staking fund every month with an initial buffer of $10,000, using the profits to continue staking. At the end of the month, they use the profits to buy back Alpha tokens. The team made about $18,000 in profits in March.

Depending on how you use the intelligence, the results can also be very different. For example, the intelligence shows 35:65, but the actual market odds may be 40:60. Someone might bet on this, but you might not because the difference is small and there is not enough advantage. Billy uses intelligence differently than Sportstensor. (No one knows how to get a consistent positive ROI yet, as it is still early days.)

Sportstensor plans to further monetize their intelligence by creating a dashboard that allows users to easily understand insights and make betting decisions based on them.

I personally like this team because there are so many directions their product can go. We have seen how Billy has captured attention and got sports fans excited to follow the betting. With the team covering multiple sports, the agency has the potential to change the way people interact, bet, and feel about betting.

SN44 @webuildscore

Score originally built a product similar to Sportstensor, but turned to computer vision after realizing that the ability to predict future events could bring more value.

To understand this, you need computer vision to analyze what’s on the screen, AI to understand objects on the screen, locate and annotate the data, then use different algorithms to draw conclusions (e.g., the probability of a player making a certain move), and convert all of this into a universal score that can be used to improve player performance (and identify talent early).

Miners compete to mark objects (this is the primary goal of miners). Score uses its internal algorithm to reach a conclusion (for now).

When you rate players (similar to Elo in chess or League of Legends, but more granular and dynamic…changing dynamically in every game based on player decisions and their impact), you can do a lot of things as a club owner, like spotting talent when players are very young. If you have videos of kids playing, you can analyze them the same way as professional games. It’s a unified way to quantify the entire world of football.

Through proprietary data, Score can monetize scores and insights, selling them to data brokers, club owners, sports data companies and bookmakers.

Soon, users will be able to upload videos on Score's self-service platform for miners to annotate. Usually, football match videos take hours to annotate, but miners can annotate 90 minutes of games in just 10-12 minutes, which is much faster than other platforms. Users can use the annotated data for their own models and use cases.

I like Score because it can be applied to areas beyond sports, such as self-driving cars, robots, etc. In a world full of data garbage, high-quality proprietary data is extremely valuable.

SN18 @zeussubnet

This is a new subnet that has been getting a lot of attention lately. I haven't had time to speak with the team yet, but the product is very interesting.

Zeus is a machine learning-based weather/climate prediction subnetwork designed to outperform traditional models and provide faster and more accurate predictions.

This intelligence is highly sought after by hedge funds because accurately predicting the weather can lead to better predictions of commodity prices (hedge funds are willing to pay millions of dollars for this intelligence because if they can win at commodity trading, they can make hundreds of millions of dollars).

The Zeus subnet is new, having recently acquired subnet 18. Its Alpha token is up 210% in the past 7 days.

Other subnets of interest but not yet explored

@404gen_ SN17 - Infrastructure for AI-generated 3D assets. Create 3D models for games, AI characters, virtual anchors, etc. Recently

· @unity integration could enable seamless 3D model generation, transforming the creative workflow for Unity’s 1.2 million monthly active users.

@metanova_labs SN68 - Decentralized Science (DeSci) drug discovery network, transforming drug discovery into a collaborative, high-speed competition, addressing traditional challenges such as cost and time (the traditional process takes more than a decade and costs billions).

There is a lot more to it, which I will share once I dig deeper. I will start with the easiest to understand (since I am not a technical person).

Summarize

I try not to get too technical. There are many good resources out there on technical explanations of dTAO, emissions, incentive allocation, and all the stakeholders involved.

Based on my experience during the proxy boom (October 2024 to present), it is important to be flexible. I have held tokens of too many projects, and I think dTAO provides a good mechanism for me to flexibly rotate different investment DeAI startups.

There are not many participants at present, and users can get 80%-150%+ APY, plus the price appreciation of subnet tokens. This dynamic may change in the next 6 months, as more people join and the bridge, wallet and transaction infrastructure of the TAO ecosystem will be better.

For now, I suggest you enjoy TAO's PvE season and join me as we learn more about cool DeAI technology.

「 Original link 」

Welcome to join the BlockBeats official community:

Telegram subscription group: https://t.me/theblockbeats

Telegram group: https://t.me/BlockBeats_App

Official Twitter account: https://twitter.com/BlockBeatsAsia