Securing local LLMs for academic research: a human-system integration analysis and evolution of TAUCHI-GPT
Bibliographic record
Abstract
Abstract The application of Large Language Models (LLMs) in academic research faces unique challenges of privacy and workflow integration. This paper introduces TAUCHI-GPT, a novel, open-source AI assistant whose evolution informs our analysis. We detail its two versions: a cloud-based V1 using GPT-4 and reflection cycles, and a local, privacy-preserving V2 with RAG architecture. Based on empirical findings from two user studies, we present a critical Human-System Integration (HSI) analysis of the security vulnerabilities and alignment challenges inherent in local LLM deployments. We examine how recent development trends—such as model distillation and reward-model learning—and the complexities of internal model mechanisms exacerbate risks like prompt injection, RAG data failures, and unfaithful explanations that impact user trust. Drawing from HCI principles and mechanistic interpretability insights, we propose and discuss a multi-layered mitigation strategy. This work contributes significantly to HSI and AI by presenting an evaluated system, a rigorous analysis of local deployment risks from a sociotechnical perspective, and actionable, stakeholder-specific guidelines for the secure and responsible utilization of LLMs in academia.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.028 | 0.076 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.002 | 0.005 |
| Scholarly communication | 0.007 | 0.009 |
| Open science | 0.003 | 0.010 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".