ByteDance initially distinguished itself in the internet age through its App factory model, characterized by agile development and a competitive horse-racing mechanism. However, upon venturing into the realm of large-scale models, these former strengths have transformed into impediments to progress. The development of large-scale models represents a precise, heavy-industry-level systems engineering endeavor that necessitates global coordination and collaboration. Nevertheless, ByteDance's horse-racing mechanism fragments computing power into isolated resource pools, fostering issues such as resource hoarding and information silos.

Simultaneously, the short-cycle OKR (Objectives and Key Results) evaluation system has dampened enthusiasm for investing in crucial foundational tasks like data cleaning. This has even resulted in internal sabotage and mutual avoidance of responsibility, culminating in incidents of training data contamination. Moreover, the R&D team lacks a stress-free scientific research environment conducive to long-term focus. Despite substantial investments in computing power, these efforts have only yielded localized engineering enhancements, making it challenging to achieve breakthroughs in underlying paradigms.

Currently, ByteDance has embarked on initiatives to consolidate team organizations and revise its evaluation mechanisms. Indeed, all traditional tech giants that persist in applying the outdated organizational logic of the mobile internet to deploy large-scale models encounter similar challenges. Only by dismantling entrenched departmental barriers and establishing a long-term scientific research and industrial system tailored to large model development can significant breakthroughs be achieved in this domain.