Heurist Mining Season 2: From POW to DePIN

Heurist
7 min readJul 16, 2024

--

As we move closer to the conclusion of Heurist Mining Season 1 where 5% of total Heurist Tokens will be allocated to 🦙 Llama and 🧚‍♀️ Waifu Points miners, it’s crucial to understand the evolving landscape of blockchain mining and how Heurist is pioneering a new approach in the realm of Decentralized Physical Infrastructure Networks (DePIN).

The incentive structures of Proof of Work (POW) blockchains and DePIN serve different purposes and face unique challenges.

POW Mining Properties

  • Predictable emission
  • Often results in oversupplied compute, leading to wasted tokens on hardware providers not serving real-world use cases

This approach uses lots of computational power and energy for the sole purpose of securing the blockchain. However, as Heurist is a DePIN protocol built upon its own Layer-2 network using ZK Stack, the security is guaranteed by Ethereum and ZK proofs. There is limited benefit of incentivizing excessive computational power without serving real-world use cases.

DePIN Mining Properties

  • Emission based on effective contribution
  • Aims to reduce idle resources to prevent token hyperinflation

Token emission in DePIN serves as an incentive for physical resource providers who actively serve users. The key distinction is that DePIN mining rewards should directly correlate with the network’s real-world usage. Otherwise, the protocol would be over/under-paying miners if the mining rewards do not reflect the actual demand of the network.

However, based on the observation of Heurist testnet mining and other DePIN protocols, there are certain meritorious properties that can never be achieved altogether. We call it the “impossible triangle” of DePIN mining:

  1. Projectable rewards: Miners can expect the rewards to grow at a predictable rate
  2. Permissionless mining: Miners can join or leave the network at any time, without whitelisting
  3. No wasted compute: The computational resources from miners contribute to real-world demands

Let’s review several example protocols to understand why these three properties do not co-exist.

Protocol A: Miners install IOT devices in their homes to provide wireless telecommunication network. It’s permissionless to join the network, and every device contribute to providing useful physical resources (in this case, wireless network). However, the rewards are not steady but rather depend on network usage, location and various factors.

Protocol B: Miners provide datacenter servers and get compensated with token rewards based on the quality, location and duration. The amount of rewards are well-defined, and the compute resources are never wasted idle. However, it’s not permissionless to join as a compute provider, but it requires off-chain legal agreements.

Protocol C: Miners rent out their CPU or GPU PCs in a peer-to-peer manner. Daily rewards are relatively steady, and it’s permissionless to join mining. However, due to speculation of token price and a lack of demand for renting consumer-grade PCs, the compute resources in the network are over-supplied.

Recap of Heurist Season 1 Mining

As with any groundbreaking project, Heurist’s testnet mining phase faced its share of challenges. We observed two primary issues during this period:

(a) The volatility in daily point earnings created uncertainty, potentially discouraging late-coming miners due to unpredictable future reward rates. (b) A significant amount of compute resources were allocated to processing bot requests rather than genuine user queries, leading to inefficiencies.

However, these imperfections were not unexpected. They are, in fact, a natural part of the innovation process, especially during a testnet phase. True innovation demands experimentation, trials, and continuous calibration. Our testnet mining served as a crucial learning experience, allowing us to identify areas for improvement and refine our approach for the future.

Despite these challenges, it’s important to take a moment to reflect on the remarkable achievements of our testnet mining phase. The numbers speak volumes about the enthusiasm and engagement of our community:

  • Over 7,000 registered miner addresses
  • A consistent pool of 3,000 to 6,000 online GPUs
  • Over 200 million image generation requests processed
  • Over 500 million text generation requests processed

This level of engagement and processing capacity is a testament to the potential of decentralized AI infrastructure and the commitment of our mining community.

With these learnings and achievements in mind, let’s review the key details of Season 1 Mining:

  • Season 1 Mining, which began on April 1st, 2024, will conclude on July 19th, 2024 (16 weeks total)
  • A final snapshot of earned Llama & Waifu Points will be taken on July 19th.
  • 2.25% (22,500,000) of total Heurist Tokens will be allocated to Stable Diffusion miners (🧚‍♀️Waifu point holders) Proportional to points earned.
  • 2.75% (27,500,000) of total Heurist Tokens will go to LLM miners (🦙Llama point holders) Proportional to points earned.
  • These tokens will be claimable 24 hours after the Heurist Token becomes tradable on CEXs and DEXs.

Note: The Token Generation Event (TGE) date is yet to be determined.

We will allocate an extra bonus pool with 0.4% token supply (4,000,000 from unallocated NFT airdrop) to reward miner addresses based on the time that they start mining.

FAQ

Q: Why is it 2.25:2.75 ratio between Waifu and Llama points instead of an even split?

A: Running LLMs requires higher-end GPUs (min 24GB VRAM). Running Stable Diffusion only requires 12GB VRAM. We have done the maths estimating the total costs of GPU-hours of all SD miners and LLM miners, and the ratio is close to 2:3. However, some miners prefer that there should be an even split between Waifu and Llama. Considering all these factors, 2.25:2.75 ratio is a middle ground that most miners are aligned with.

Q: Why is there 24 hours delay in the claim time?

A: Market makers typically need some time after TGE to start working. We want to allow sufficient buffer time to ensure there’s enough liquidity to support trading after TGE.

Q: Why is there a non-linear bonus pool?

A: Early miners have taken higher risks and spent more time on configuring and troubleshooting early versions of mining software. Such efforts need to be compensated. Additionally, we’d like to reward node rental users for the trust that they have towards Heurist Team.

Introducing Season 2

Season 2 mining will commence immediately after Season 1 ends on July 19th. Keeping the “impossible triangle” of DePIN mining in mind, we have designed a plan to improve the clarity of mining emissions by balancing the three angles (a) Future rewards should be projectable to a certain degree (b) Mining will remain permissionless (c) We aim to minimize the token emission wasted on rewarding the processing of bot requests.

Here’s what’s new:

  1. Dynamic Emission Rates

Waifu Pool (Stable Diffusion)

  • Base rewards: 200,000 tokens (2 bps of total supply) weekly
  • Dynamic rewards: Up to 500,000 tokens (5 bps of total supply) weekly, based on the usage of Stable Diffusion models

Llama Pool (LLM)

  • Base rewards: 200,000 tokens weekly
  • Dynamic rewards: Up to 500,000 tokens weekly, based on the usage of LLMs

The base rewards ensure that miners are guaranteed to earn rewards regardless of network usage. This ensures a baseline level of compute resources in the network. The dynamic rewards will be determined by organic (non-bot) demand for AI inference compute. We’ll provide dashboards to display this data collected from our inference APIs. This allows the compute power of Heurist network to scale up as more compute demand is onboarded.

2. Weekly Snapshots

Unlike Season 1’s single end-of-season snapshot, Season 2 will feature weekly snapshots. This means Waifu and Llama points in Season 2 will be converted to Heurist Token balances every week. This provides clear reward visibility, prevents point hyperinflation, and helps miners to project their rewards.

3. Token Claim Timeline

Season 2 rewards will become claimable as liquid Heurist Tokens in Q4 2024 or Q1 2025, depending on the mainnet launch time of ZK Stack and Heurist ZK Chain. We understand that it might take a longer than predicted time to build, test, and audit blockchain infrastructures. Therefore, we will make Season 2 rewards claimable when (1) Heurist ZK Chain goes live (2) 3 months after TGE, whichever comes first.

4. Mining Multiplier Rules

To further incentivize participation and align the interests of various stakeholders in our ecosystem, we’re introducing mining multiplier rules:

Pre-TGE: Heurist Imaginaries NFT holders will benefit from mining multipliers, as defined in the snapshot proposal: https://snapshot.org/#/heurist.eth/proposal/0x508c8441c2fd2c33b717489fdbd8cba9df55666d6a42310d7d837f2f5ae9a2e2

Post-TGE: Once the Heurist Token is live, the NFT multipliers will be disabled. Instead, the multiplier will be determined by the miner’s staked Heurist token (stHEU token) balance. This shift encourages long-term commitment to the Heurist ecosystem and aligns mining rewards with token holders’ interests.

Beyond Season 2: The complete set of staking, voting and bribing mechanisms as described in https://docs.heurist.ai/overview/tokenomics will be implemented on Heurist Chain.

Summary of the differences between Season 1 and Season 2

Get Involved

Whether you’re a seasoned miner from Season 1 or new to the Heurist network, now is an excellent time to get involved. By contributing your computing resources to our decentralized AI inference protocol, you’re helping to democratize AI and bridge the worlds of crypto and artificial intelligence.

For the latest information and to start mining, visit our documentations or join our Discord.

--

--

Heurist

DePIN of serverless GPUs powering open source AI innovations