This is part of our series on a year of Bittensor experience, leading up to our anniversary at the 13th of July. We discuss 9 miner archetypes before digging deeper into incentive landscapes. After we’ve discussed honest, rogue, calculative, blockchain-savvy, overfitting, devops and (over)analyzing miners, a good one to discuss is the clueless miner. Please let us know what you think in our Discord channel!
Some miners haven’t got the slightest clue about any of the things we’ve written in this series. They come in on Discord, ask for help on installing the official miner code, then spend their Tao on registration fees, earn nothing, and get deregistered.
We should discourage these miners from participating, by preferably not publishing miner code they can only use to hurt themselves. Successful miners invariably write their own miner code. People that can’t figure out how to do that, are probably in the wrong business.
This doesn’t mean subnet owners should not publish helpful tools, such as code to evaluate and post results, or provide a basic framework for a miner. The SN29 (LLM training) validator code for example was explicitly designed to allow local model injection, so that miners could benchmark their models offline, against winning models, before uploading them. The upload script ensured that no miner wasted time on getting the submission format and embedded hashes right.
A particularly sad case of clueless miners are those that spend significant amounts of real money on GPU training runs, only to see their result being surpassed in minutes after they are submitted. Full disclosure: we’ve been on both ends of this. We all started out as clueless miners.
Next up we have a slightly more successful type: the slow miner.
0 Comments