Predatory journals and paper mills have been an ominous presence in academic publishing for decades. More recently, GenAI (generative artificial intelligence) has made an impact, accelerating the assault on the integrity of academic publishing. This article surveys these phenomena partly from a journalistic angle, tracing their rapid ascent and the destruction they continue to cause. 

The nature of language, its representation in the mind of humans, and how it is processed and eventually acquired should constitute the basic knowledge in the development of an effective and evidence-based language teacher-education programme. The lack of language experts currently teaching in these programmes has often led to the perpetuation of misleading beliefs and wrong assumptions about language development and language instruction. For example, the old belief that language is a list of rules, such as those found in textbooks, and that ‘knowing a language involves knowing its rules’ has led to the misleading view that teaching grammar explicitly is necessary or even beneficial for language development. The main consequence of perpetuating misleading beliefs is one hand the development of ineffective language teacher education programmes which are not evidence-based, and on the other hand the development of misleading assumptions for language instruction. Language teacher education programmes are often about training language instructors to use textbooks and perpetuating the use of traditional language teaching methodologies. In this paper, the mismatch between misleading beliefs, assumptions and facts based on scientific evidence in language development are highlighted with the view of providing an effective way forward in the development of appropriate language teacher education programmes. 

Integrating Generative AI into English for Academic Purposes (EAP) risks inducing “cognitive offloading,” where thinking processes are outsourced to AI, undermining the critical literacy engineering students need to analyze complex Fintech issues. This two-year action research study (2024–2025) at the Chinese University of Hong Kong (N=96) evaluated whether constraint-based AI protocols can successfully shift AI from a simple content generator to a metacognitive scaffold. Using mixed methods (surveys and interviews), the study examined an intervention featuring a “Prompt-Observe-Evaluate” reading protocol and “Persona-Based” writing feedback, explicitly banning direct AI summarization. Results showed significantly increased student self-efficacy in ethical AI use. Students also reported improved critical thinking, driven by the requirement to verify AI outputs against source texts to spot hallucinations and bias. By 2025, 100% of participants recognized ethical risks like dependency and data privacy. Ultimately, these findings suggest that EAP instruction requires “human-in-the-loop” constraints, positioning AI as a collaborative reasoning partner rather than a substitute for active cognitive engagement.

The application of Generative Artificial Intelligence (GenAI) in various writing contexts has transformed the writing process from a sheer manual cognitive task into human-AI collaborative writing. This paradigm shift necessitates a reconceptualization of writing constructs within higher education to ensure that writing assessments accurately reflect the evolving knowledge and skills demanded in the authentic AI-assisted writing context. Grounded in the argument-based validity framework, this qualitative study investigates the domain definition inference of three university English writing courses in Hong Kong. Data collection involved a comprehensive analysis of construct artifacts, including course outlines, teaching materials, and assessment task specifications, supplemented by semi-structured interviews with instructors. The data were mapped against Cardon et al. (2023)’s AI literacy framework comprising four dimensions: application, authenticity, accountability, and agency. Findings reveal significant discrepancies in operationalization of AI literacy across the faculties. All courses exhibited construct under-representation in the application dimension due to a lack of instruction on iterative prompting and operational mechanisms of Large Language Models (LLMs). One case even demonstrated construct irrelevance by utilizing AI detection reports that inhibit human-AI collaboration. New assessment specifications are proposed in the end to bridge the gap between conventional instructional/assessment design and the requirements for effective AI-mediated writing.