Large Language Models (LLMs) are increasingly deployed in tools for thought and creativity support systems, yet their generative fluency can collapse early-stage ideation into selection and curation, shifting cognitive labor away from concept formation and divergent thinking toward homogenized outputs. We introduce concept suppression as a subtractive approach to induce cognitive friction, treating the learned conceptual manifold of a pre-trained model as a design surface for human-AI interaction. We operationalize this with Unlearning to Rest, a prototype that applies weight-level concept suppression to Llama3.2:3b, suppressing a canonical attractor concept ("chair") to create navigational impediments in the solution space. We evaluate the approach in a within-subject study (N = 37) where participants were set a conceptual ideation challenge. Quantitative results show a consistent workload-ownership trade-off: the unmodified model reduces mental demand and effort but also reduces perceived ownership, while Unlearning to Rest yields ratings closer to unaided ideation. Qualitative analysis indicates that the constrained model prompts user re-articulation, supporting a stage-fit workflow in which constrained assistance benefits early ideation. We contribute (1) representational intervention as a lens for LLM-based tools for thought, (2) a system instantiation via weight-level concept suppression, and (3) empirical evidence motivating stage-sensitive evaluation of LLM assistance beyond efficiency and output quality.
Generative AI (GenAI) systems are transforming cognitive work, yet we lack a clear picture of how different human-AI interaction structures affect the output participants produce, the process they go through, and how they perceive the AI partner. We conducted a between-subjects experiment (N = 244) comparing four implemented interaction designs for a user persona creation task: human-lead expand, AI-lead propose, iterative critique, and free dialogue. Iterative critique produced higher task-appropriate creativity than AI-lead propose, driven by appropriateness and the composite creativity score; free dialogue and human-lead expand produced more descriptively rich language than iterative critique; originality did not differ across conditions. A between-persona similarity analysis found that interaction structure converges iterative critique participants on a common persona, while AI-lead propose diversifies scenarios relative to chance. Participants in the condition producing the highest task-appropriate creativity perceived the AI as a tool and felt personal ownership, while those in more conversational conditions perceived the AI as a collaborative partner. We conclude that interaction pattern choice should be driven by what outcome and experience the designer is aiming for, and we offer preliminary recommendations for matching interaction patterns to those goals, including the possibility of pattern switching across phases of a single workflow.
In Computer-Supported Collaborative Learning (CSCL), Group Awareness Tools (GATs) collect raw data produced during collaboration, interpret awareness information from that data, and visualize awareness information to support students in perceiving differences among themselves in cognitive and social states. These differences can trigger coordination, which in turn supports meaningful learning. Given its language understanding and generative capabilities, Generative AI (GenAI) holds the potential to enrich the awareness of information interpreted from raw data by traditional GATs. To explore this potential, we conducted a Systematic Literature Review (SLR) of 689 papers and synthesized findings from 32 GAT studies. From this, we derived analytical hypotheses about what awareness information GenAI could enrich, which UI and interaction techniques could be used to present and explore this enriched awareness information to potentially prompt coordination, and how different UIs and interaction techniques could be selected for different types of awareness information. This work provides a starting point for designers to design GenAI-based GATs that could support collaborative learning.
The excessive excretion of antibiotics in wastewater poses significant environmental risks. Adsorption is expected as highly efficient, cost-effective, and sustainable methods for treating these contaminants. This study investigated the adsorption of tetracycline hydrochloride (TC) on palm shell derived activated carbon (PSAC) using the advanced pore volume and surface diffusion model (PVSDM). Scanning electron microscopy and N2 adsorption isotherm revealed a highly microporous PSAC with a specific surface area of 852.85 m2 g-1, an average pore diameter of 2.068 nm and a micropore proportion of 0.72. In the developing PVSDM, conventional isotherms (e.g. Langmuir, Freundlich and Redlich-Peterson) and global kinetics models (e.g. pseudo-first order, pseudo-second order and Elovich) were comprehensively analyzed and the most suitable isotherm and global kinetics models were incorporated into the PVSDM. External film diffusion and intraparticle diffusion (IPD) investigation, according to Weber-Morris and Boyd, consistently confirmed strong IPD control. The PVSDM effectively reproduced time-dependent adsorption capacity in batch experiment and was verified by reasonable prediction of the dynamic change of adsorption capacity within the PSAC particle. The determined mass transfer parameters indicated that pore volume diffusion was the dominant mechanism, while surface diffusion was insignificant. This study demonstrates the effectiveness of PVSDM for quantifying mass transfer resistances in adsorption of TC on PSAC, providing valuable insights for the design and optimization of large-scale adsorption equipment for treatment of antibiotics in wastewater.
Acetone is categorized as a volatile organic compound (VOC) gas. Exposure to it for a long period harms human health and the environment. Commonly, a single sensor is used to sense the target gas. However, it always suffered from low selectivity, making it difficult to analyze and differentiate various gases. To overcome this problem, a gas sensor array is proposed in this work to investigate the response value for each array to the acetone vapor. Six sensor arrays based on various ratios of TiO2 doped with graphene were deposited using the screen-printing technique on an FR3-printed circuit board (PCB). The gas sensor arrays were exposed to two different concentrations of acetone vapor at room temperature. Next, the response values were classified using a deep neural network (DNN). The results revealed that the T95_G5 gas sensor responds more to 25 mL and 35 mL of acetone gas, with 1.0966 and 1.2074, respectively. The overall DNN accuracy for acetone vapor was 82.70%.
