Quasar's 120B AI Model Scandal: Decentralized Training Provenance Under Scrutiny

Decentralized AI project Quasar released what it called a groundbreaking 120 billion-parameter model, touted as a triumph of decentralized training on the Bittensor network.

The model, marketed as a key innovation featuring architectural innovations like the 'Loop Transformer' and 'Engram memory', was supposed to solve memory and coherence problems plaguing standard transformer architectures.

However, independent analysts found that roughly 96-98% of its weights closely matched Ant Group's Ling-mini-base-2.0 model, sparking a swift market backlash.

The SN24 alpha token dropped approximately 70-75% in a single trading session, underscoring the crypto community's scrutiny over training provenance and architectural claims.