Quasar recently unveiled a 120 billion-parameter AI model trained on the Bittensor subnet 24, aiming for decentralized training. However, the alpha token price has plummeted approximately 70% following an analysis that found about 98% of the model’s weights matched an existing open model, raising concerns about the extent of original training versus inherited parameters. This incident underscores the increasing demand for verifiable training provenance in decentralized AI, as such transparency is becoming a crucial requirement to establish the credibility of AI model development and meet regulatory expectations for institutions.

Quasar: Quasar runs as Subnet 24 on the Bittensor network and develops a decentralized marketplace for training long-context large language models through coordinated miners and validators. The project positions its subnet framework as a way to achieve distributed model training without centralized infrastructure. In the news, Quasar claimed decentralized training of a large-parameter model on subnet 24, prompting scrutiny over training provenance.
Bittensor: Bittensor is a decentralized blockchain network that creates open markets for machine intelligence by coordinating miners who generate AI outputs and validators who score them across specialized subnets. Participants earn incentives through the network’s token for useful contributions to AI tasks such as inference and training. The news highlights activity on one of its subnets, illustrating Bittensor’s role in enabling claims of decentralized AI model development.
Mark Jeffrey: Mark Jeffrey is a Bittensor-focused investor, author of books on Bitcoin, and partner at Stillcore Capital, with expertise in subnet tokens and decentralized AI. He was recently appointed advisor to TAO Synergies to guide strategy in the Bittensor ecosystem. In the news, he publicly shared and endorsed the Quasar subnet’s model training claim on X.

Ledger Utility: Decentralized immutable ledgers are well suited to record and verify AI training processes and agent activities.
Verifiable Training: Verifiable training provenance is emerging as a core requirement for decentralized AI projects to establish credibility of model development.
Institutional Requirements: Verifiable agent logs are becoming a regulatory expectation for institutions deploying AI systems.