Recent Security Breach Attempts and Fine-Tuning Claims Against NovaSpark (iqlvbs.com)

IQ's NovaSpark driving model has faced accusations from Comma's Harald Schaefer, who claimed it is merely a short fine-tuned version of their Rebellious Hope model. In response, NovaSpark's developers assert that their model was trained independently, highlighting that a detailed parameter analysis indicates zero correlation between NovaSpark and Rebellious Hope. Key differences were revealed: NovaSpark has approximately 2.2 times the number of parameters and significantly higher computational requirements, emphasizing its unique architecture, including more vision blocks and altered normalization weights. The differences in model outputs and internal structures refute Comma's claims of similarity. Moreover, NovaSpark's team reported multiple security breach attempts aimed at stealing their model and accessing user data, with identifiable requests originating from a Comma.ai IP address. These attempts included evading model-access restrictions, bypassing ownership checks, and exploiting sign-in protocols, all aimed at gaining unauthorized access. Despite the breaches being thwarted, the incident raises significant concerns about security in the AI/ML community, particularly surrounding proprietary models and user privacy. The situation underscores the importance of robust security measures as AI technologies evolve and inter-company conflicts arise.