5월 5주차

When should a game studio continue building in beta and when should it go live? In an industry defined by continuous updates and diverse players, online game developers must decide what to build and when to release while balancing engagement, monetization, and development resources. Unlike traditional software, online games operate in a mass market environment where users differ widely in experience and tolerance for bugs, and where frequent updates are essential to maintain interest and competitiveness. We show that beta releases are not only testing tools but strategic instruments for managing user diversity, product complexity, and release timing. Beta phases allow firms to gather feedback and refine features before broader release. Full releases should be delayed when complexity is high or when experienced users represent a large share of the user base. Release decisions must balance development constraints with market pressures. Although delaying releases can improve quality, competition may require earlier launches. From a policy perspective, platform operators can support better outcomes by enabling structured beta programs, managing user expectations, and providing tools for staged rollouts. These practices help improve product quality, reduce negative user experiences, and enhance long-term user engagement in digital ecosystems.

Information privacy concerns shape consumer behaviors, regulatory choices, and the design of AI-driven products, yet prior research offers contradictory guidance. We synthesize three decades of evidence from 305 empirical studies, mapping the antecedents, consequences, and moderators of privacy concerns. For practitioners, the strongest levers are situational rather than demographic: assurance mechanisms, perceived control, information sensitivity, and perceived vulnerability have stronger effects than age, gender, or personality. Resultantly, systems and regulations that enhance user control and transparency outperform demographics- or personality-based approaches. However, context shapes these relationships. Cultural and regulatory environments, the type of risk (financial, social, physical, or general), and whether a system relies on conventional data flows or AI-driven inference moderate how users respond. Notably, AI-augmented systems invert the classic privacy calculus: concerned users often keep sharing, suggesting that trust-building and assurance mechanisms differ for new analytics and AI systems. For policy, our findings support investments in meaningful transparency, standardized assurance disclosures, and risk-tailored safeguards, particularly for AI governance. For research, we identify underexplored but high-impact variables, including privacy empowerment, psychological ownership, app permission concerns, and regulatory awareness. These variables should anchor the next generation of privacy studies.

Large language models (LLMs), such as ChatGPT, exhibit substantial functional overlap with user-generated knowledge ecosystems while also relying on them as critical inputs for future learning. This dual role creates a fundamental tension that calls for a clearer understanding of how LLMs reshape these ecosystems. Adopting a niche theory perspective, we examine how functional overlap and knowledge structure determine the boundary between substitution and coexistence. Using Stack Overflow, we show that LLM introduction reduces question volume by about 14% on average (and up to 27.9% over time), with stronger declines in mid- to low-quality content, in topics with richer and more structured knowledge bases, and among less experienced users. Conditional on similar question activity, topics with deeper answer-side knowledge experience disproportionately larger reductions, highlighting the role of accumulated knowledge. These patterns reflect selective substitution; routine and well-documented queries migrate to LLMs, whereas complex, context-dependent problems remain. We also document direct improvements in question quality, suggesting positive spillovers from reduced search and articulation costs. Together, the findings indicate niche partitioning rather than full displacement, with a self-reinforcing knowledge flywheel between LLMs and platforms. For practice, platforms should reposition toward high-expertise niches by integrating artificial intelligence (AI)-assisted scaffolding, strengthening expert incentives, and curating complex knowledge. For LLM development, the results point toward deeper integration, where AI systems complement community knowledge production and enable more advanced problem solving.