On April 1, almost a week after maintaining its silence, the team responsible for Google's contentious TurboQuant compression algorithm paper finally issued a response. However, this move did little to quell the escalating controversy. Addressing the accusations of striking technical parallels, Google asserted that random rotation is a widely recognized technique and contended that any inaccuracies in the experimental benchmarks had minimal bearing on the core findings. Earlier, the paper had touted that the TurboQuant algorithm could slash the memory requirements of large model KV caches by at least six times, amplify speeds by up to eightfold, all without compromising accuracy. This revelation prompted a dramatic slump in global memory chip stocks, with market valuations plummeting by over $90 billion. Yet, Jianyang Gao, a postdoctoral researcher at ETH Zurich, leveled serious allegations against the Google team, citing systemic academic misconduct. This included evading similarities with the RaBitQ method, misrepresenting theoretical outcomes, and conducting experiments under biased conditions. Google's rebuttal failed to alleviate these concerns, and the controversy continues to intensify.
