Quasar, a decentralized AI project operating on Subnet 24 of the Bittensor network, unveiled a 120 billion-parameter model with a 5 million-token context window. Within hours, independent researchers found that 96 to 98 percent of the model’s weights were nearly identical to Ant Group’s Ling-mini-base-2.0 model. The SN24 alpha token crashed approximately 70 to 75 percent in a single trading session. SILX AI, the project developer, had marketed architectural innovations called « Loop Transformer » and « Engram memory », which were largely absent from the released model.
Source: Read the original article
