Manuel B. Garcia

Manuel B. Garcia serves as the Senior Director for Educational Technology and Digital Learning at FEU Institute of Technology, Manila, Philippines. Read More

Contact Info

1607, FEU Tech Building,
P. Paredes St, Sampaloc,
Manila, Philippines
mbgarcia@feutech.edu.ph

Follow Me

The Complete Library

All Research Guides

Browse every published guide in the Researcher's Field Guide, including the complete research journey and supplemental collections, listed alphabetically for quick reference.

Complete Library

Every published guide

Showing 101–150 of 5,065 guides

5,065 guides
PART 0D · SUBPART 83 · GUIDE 1096

Scientific Justification for Large AI Models

Is Using the Largest Available AI Model Always Scientifically Justified?

The largest available AI model is not necessarily the most appropriate research instrument. Learn how to determine whether additional model capabilities provide meaningful scientific benefits, when smaller alternatives may suffice, and how to justify computationally demanding choices.

PART 0D · SUBPART 83 · GUIDE 1095

Smaller AI Models for Research

Can Smaller AI Models Be Scientifically Sufficient for Some Research Tasks?

Smaller AI models may be scientifically sufficient for classification, information extraction, and other well-defined research tasks. Learn how to establish adequacy through task-specific validation, examine consequential errors, and determine when a larger model is genuinely necessary.

PART 0D · SUBPART 83 · GUIDE 1094

Reporting Computational Resources in AI Research

Should Large-Scale AI Research Report Computational Resource Use?

Large-scale AI research can require substantial computing infrastructure, yet publications sometimes provide limited information about the resources consumed. Learn which computational measurements matter, how to report them, and why transparency supports reproducibility without requiring every study to disclose identical metrics.

PART 0D · SUBPART 83 · GUIDE 1093

Energy Use in AI Research Methods

Should Researchers Consider Energy Use When Choosing AI Methods?

Energy consumption can be a legitimate consideration when selecting AI research methods, particularly for large computational workloads. Learn how to compare energy requirements without sacrificing scientific validity, adequate validation, or reproducibility.

PART 0D · SUBPART 83 · GUIDE 1092

Environmental Impact of AI Queries

Does Every AI Query Have the Same Environmental Impact?

AI queries do not all consume the same amount of energy or have identical environmental impacts. Learn how model architecture, processing demands, hardware, and electricity sources affect environmental estimates, and why published per-prompt figures require careful interpretation.

PART 0D · SUBPART 83 · GUIDE 1091

Environmental Cost of AI Research

Does Using AI for Research Have an Environmental Cost?

AI-assisted research has an environmental footprint because models depend on electricity, computing equipment, and supporting infrastructure. Learn where these impacts arise, why they vary, and what researchers can reasonably measure without relying on misleading per-query estimates.

PART 0D · SUBPART 82 · GUIDE 1090

Choosing AI Deployment Options for Research

How Should Researchers Choose Between Public Cloud, Enterprise, Institutional, and Local AI?

Public cloud, enterprise, institutional, and local AI offer different combinations of control, cost, security, and computing capacity. Learn how to compare these deployment arrangements and select one that meets your research requirements without assuming any option is universally superior.

PART 0D · SUBPART 82 · GUIDE 1089

Institutionally Hosted AI for Research

Can Institutionally Hosted AI Provide a Useful Middle Ground for Researchers?

Institutionally hosted AI can offer researchers shared computing resources, centrally managed security, and greater control over model deployment. Learn when this arrangement provides a practical alternative to personal local installations and commercial cloud services.

PART 0D · SUBPART 82 · GUIDE 1088

When Local AI Is Impractical

When Is Running AI Locally Impractical for a Research Team?

Local AI offers greater control over data and model versions, but it is not always practical. Learn when hardware constraints, maintenance, security responsibilities, or research demands make cloud or institutionally hosted alternatives more defensible.

PART 0D · SUBPART 82 · GUIDE 1087

Hardware Requirements for Local AI Models

What Hardware Do Researchers Need to Run AI Models Locally?

Running AI locally does not always require an expensive GPU, but hardware needs vary considerably by model and workload. Learn how to estimate memory, processing, and storage requirements before investing in local AI infrastructure.

PART 0D · SUBPART 82 · GUIDE 1086

Validating Local vs. Cloud AI Models

Can a Local AI Model Be Validated More Easily Than a Changing Cloud AI Model?

Local AI models can offer greater control over model versions and testing conditions, while cloud models may change without researchers controlling every component. Learn how to validate each arrangement, detect consequential performance changes, and preserve methodological comparability.

PART 0D · SUBPART 82 · GUIDE 1084

Open-Weight AI and Research Reproducibility

Can Open-Weight AI Models Improve Research Reproducibility?

Open-weight AI models can make research easier to reproduce by allowing researchers to preserve model parameters and control inference environments. Learn what this improves, why identical outputs are not guaranteed, and which additional artifacts are needed for credible computational reproducibility.

PART 0D · SUBPART 82 · GUIDE 1083

Local AI and Research Confidentiality Risks

Does Running AI Locally Eliminate All Research Confidentiality Risks?

Local AI can keep research data away from external inference providers, but it does not eliminate confidentiality risks. Learn how local systems can still expose sensitive information and what researchers should verify before processing confidential data.

PART 0D · SUBPART 82 · GUIDE 1082

Privacy Advantages of Local AI Models

What Are the Privacy Advantages of Running an AI Model Locally?

Running AI locally can reduce external data transfers and give researchers greater control over confidential information, storage, and processing. Learn which privacy advantages are meaningful, what safeguards remain necessary, and why local deployment does not eliminate every confidentiality risk.

PART 0D · SUBPART 82 · GUIDE 1081

Local vs. Cloud AI Models for Research

Should Researchers Use Local AI Models Instead of Cloud-Based AI?

Local AI models offer greater control over computing environments and data processing, while cloud services can provide convenient access to advanced capabilities. Learn how to choose between them based on scientific requirements, confidentiality, costs, technical capacity, and reproducibility.

PART 0D · SUBPART 81 · GUIDE 1080

Environmental Costs of AI at Scale

Should Researchers Consider Environmental and Resource Costs When Using AI at Scale?

Large-scale AI research consumes electricity, computing capacity, and infrastructure resources. Learn when environmental costs become methodologically relevant, how to evaluate them, and how to reduce unnecessary resource use without weakening scientific evidence.

PART 0D · SUBPART 81 · GUIDE 1079

Reporting AI Research Computational Costs

How Should Researchers Report the Computational Cost of AI-Assisted Research?

AI-assisted research can involve API charges, computing resources, repeated processing, and human validation. Learn what costs to document, how to calculate them, and how to report them transparently without confusing financial expenditure with computational consumption.

PART 0D · SUBPART 81 · GUIDE 1077

Paid AI Access and Research Inequality

Can Paid AI Access Create Inequality Between Research Institutions?

Paid AI subscriptions, API charges, and institutional infrastructure can create unequal research opportunities. Learn how these disparities affect research capacity, collaboration, and scientific participation.

PART 0D · SUBPART 81 · GUIDE 1076

AI Model Costs and Research Reproducibility

Can Expensive AI Models Create Reproducibility Barriers for Other Researchers?

Expensive AI models can make it difficult for other researchers to repeat computational procedures or independently test published findings. Learn how financial barriers interact with model access, documentation, and infrastructure, and what researchers can do to make AI-assisted research more reproducible.

PART 0D · SUBPART 81 · GUIDE 1075

Comparing AI Models by Cost and Performance

Should Cost Be Considered When Comparing AI Models for Research?

The most expensive AI model is not necessarily the most suitable for a research task, and the cheapest may require more corrections. Learn how to compare AI systems using scientifically acceptable performance and the total resources required to obtain usable results.

PART 0D · SUBPART 81 · GUIDE 1074

Estimating AI Processing Costs Before Research

How Do Researchers Estimate AI Processing Costs Before a Study Begins?

Accurate AI cost estimates begin with a representative pilot and a clearly defined processing workflow. Learn how to calculate expected API usage, account for repeated processing and human verification, and develop defensible cost scenarios before committing research resources.

PART 0D · SUBPART 81 · GUIDE 1073

Unexpected AI Costs in Large-Scale Research

Can AI Costs Grow Unexpectedly During Large-Scale Research Analysis?

AI processing costs may increase unexpectedly when research datasets grow, prompts become longer, or workflows require repeated analysis and verification. Learn how to identify cost escalation risks and maintain financial control without compromising research quality.

PART 0D · SUBPART 81 · GUIDE 1072

Budgeting for AI API Costs

Should AI API Costs Be Included in a Research Budget?

AI API charges should be budgeted when they are necessary for planned research activities. Learn how to estimate billable usage, justify expenses in funding proposals, distinguish API charges from subscriptions, and manage spending throughout a study.

PART 0D · SUBPART 81 · GUIDE 1071

The Real Cost of AI in Research

How Much Can AI Use Actually Cost a Research Project?

The cost of AI-assisted research extends beyond subscriptions and API charges. Learn how to estimate the total expense of using AI in a research project, including human verification, infrastructure, repeated processing, and the costs of maintaining scientific reliability.

PART 0D · SUBPART 80 · GUIDE 1068

Research Without AI Model Version Locking

What Should Researchers Do if an AI Provider Does Not Allow Model Version Locking?

Not every AI provider allows researchers to select or preserve a fixed model version. Learn how to manage this limitation through controlled workflows, systematic records, performance monitoring, and research designs that remain interpretable when the underlying system changes.

PART 0D · SUBPART 80 · GUIDE 1067

AI Model Version Locking in Research

Should an AI Research Study Lock to a Specific Model Version When Possible?

Locking an AI study to a specific model version can reduce unexpected changes in research measurements and improve methodological traceability. Learn when version locking is justified, what it actually controls, and why it cannot guarantee complete reproducibility.

PART 0D · SUBPART 80 · GUIDE 1066

Testing AI Model Update Performance Changes

How Do Researchers Test Whether an AI Model Update Materially Changed Performance?

Different outputs after an AI model update do not necessarily mean performance has improved or deteriorated. Researchers need controlled comparisons, appropriate reference standards, and meaningful performance criteria to determine whether an update matters scientifically.

PART 0D · SUBPART 80 · GUIDE 1065

Combining AI Outputs Across Model Updates

Can Researchers Combine AI Outputs Collected Before and After a Model Update?

Combining AI-generated data across model versions may introduce measurement inconsistencies that affect research conclusions. Learn when pooling is defensible, how to assess equivalence, and what alternatives are available when outputs differ.

PART 0D · SUBPART 80 · GUIDE 1064

Pausing Research After AI Model Updates

Should Research Data Collection Be Paused After a Major AI Model Update?

A major AI model update can affect research measurements, participant-related decisions, or data consistency. Learn when pausing affected research activities is justified, what to assess during a pause, and how to determine whether data collection can safely resume.

PART 0D · SUBPART 80 · GUIDE 1063

AI Model Updates as Research Instrument Changes

Should Researchers Treat an AI Model Update as a Change in the Research Instrument?

An AI model update may change how research observations are classified, scored, or interpreted, even when the study's procedures appear unchanged. Learn when an update should be treated as an instrument modification and how to assess its methodological consequences.

PART 0D · SUBPART 80 · GUIDE 1062

AI Product Behavior After Model Updates

Can the Same Named AI Product Behave Differently After a Model Update?

An AI product can retain the same name and interface while its underlying model or supporting systems change. Researchers need to distinguish genuine changes in system behavior from ordinary output variability and understand what those differences mean for research reliability.

PART 0D · SUBPART 80 · GUIDE 1061

AI Model Changes During Research

What Should Researchers Do When an AI Model Changes During a Study?

An AI model update during a study can introduce an unexpected change to the research instrument or analytical workflow. Researchers should document the update, assess its consequences, and determine whether data collection or analysis procedures require modification.

PART 0D · SUBPART 79 · GUIDE 1060

Citing AI System Performance Claims

How Should Researchers Cite Performance Claims About Rapidly Changing AI Systems?

AI performance claims can become misleading when researchers cite results without identifying the evaluated model, version, task, or testing conditions. Learn how to report these claims accurately while preserving their scientific context and limitations.

PART 0D · SUBPART 79 · GUIDE 1059

How Quickly AI Research Evidence Becomes Outdated

Can Research Evidence About AI Become Outdated Unusually Quickly?

Research findings about AI systems may lose practical relevance when models, interfaces, or deployment conditions change. Understanding what has changed helps researchers determine whether earlier evidence remains applicable or requires reassessment.

PART 0D · SUBPART 79 · GUIDE 1058

Independent vs. Vendor AI Evaluation

Why Does Independent Evaluation Matter More Than Vendor Evaluation of an AI Research Tool?

Vendor evaluations can provide valuable technical evidence, but they may not offer sufficient independent scrutiny of an AI tool's limitations. Learn how independent evaluation improves confidence in performance claims and when both forms of evidence should inform research decisions.

PART 0D · SUBPART 79 · GUIDE 1057

External Validation of AI Research Tools

How Important Is External Validation of an AI Research Tool?

Strong performance in development testing does not establish that an AI research tool will perform reliably elsewhere. External validation examines whether its performance generalizes to new data, settings, populations, and research workflows.

PART 0D · SUBPART 79 · GUIDE 1056

AI Benchmark Performance and Real-World Research

Does High AI Benchmark Performance Mean the Tool Will Work in Your Research Context?

High AI benchmark scores do not necessarily mean a tool will perform reliably in a particular research setting. Learn why differences in research tasks, disciplines, languages, data quality, and workflow conditions can undermine performance transfer, and how researchers can conduct local validation before adopting AI.

PART 0D · SUBPART 79 · GUIDE 1055

AI Training Exposure to Benchmark Questions

Can an AI Model Have Seen Benchmark Test Questions During Training?

An AI model may have encountered publicly available benchmark questions, answers, or related solutions during training. However, possible exposure is not proof of memorization or inflated performance. Learn how benchmark material enters training datasets, what evidence can establish exposure, and how researchers can evaluate model performance on genuinely unfamiliar questions.

PART 0D · SUBPART 79 · GUIDE 1054

Benchmark Contamination in AI Research

What Is Benchmark Contamination in AI Research?

Benchmark contamination occurs when evaluation material influences AI model development in ways that compromise independent performance testing. Learn how contamination can arise through training data, benchmark exposure, and repeated optimization, why it may inflate apparent capabilities, and how researchers can evaluate and reduce contamination risks.

PART 0D · SUBPART 79 · GUIDE 1053

Evaluating AI Benchmark Reliability

Should Researchers Trust AI Benchmark Results at Face Value?

AI benchmark scores can provide valuable evidence about model performance, but they should not be accepted without examining the evaluation design. Learn how benchmark contamination, task selection, scoring methods, statistical uncertainty, and changing model versions affect the interpretation of AI performance claims in research.

PART 0D · SUBPART 79 · GUIDE 1052

Evaluating AI Vendor Demonstrations

Does a Vendor Demonstration Count as Evidence That an AI Research Tool Works?

An impressive AI vendor demonstration may establish that a system can perform a particular task under selected conditions, but it does not prove consistent reliability. Learn how to evaluate demonstrations, identify missing performance evidence, ask meaningful questions, and decide when independent testing is necessary before adopting an AI research tool.

PART 0D · SUBPART 79 · GUIDE 1051

Evaluating AI Research Capability Claims

How Should Researchers Evaluate Claims About AI’s Research Capabilities?

AI research tools are frequently described as capable of reviewing literature, identifying research gaps, analyzing data, and supporting scientific reasoning. Learn how to distinguish demonstrated performance from marketing claims, evaluate the quality of supporting evidence, and determine whether an AI system is suitable for a particular research task.

PART 0D · SUBPART 78 · GUIDE 1050

Research Integrity in the Era of Cheap AI-Generated Fraud

How Should Research-Integrity Systems Adapt When Fraudulent Research Content Becomes Cheap to Generate?

Generative AI may reduce the cost of producing plausible but fraudulent research content, creating new challenges for journals and research institutions. Learn how research-integrity systems can adapt through risk-based screening, stronger evidence provenance, cross-publisher cooperation, and accountable human oversight without unfairly burdening legitimate researchers.

PART 0D · SUBPART 78 · GUIDE 1049

Unusual Academic Language and AI Detection

Can Unusual Academic Language Alone Prove That a Paper Was AI-Generated?

Unusual terminology, repetitive phrasing, and awkward academic expressions may raise questions about how a manuscript was prepared, but they cannot independently prove AI generation. Learn how editors can distinguish linguistic anomalies from evidence of unauthorized AI assistance while avoiding unfair accusations against legitimate researchers.

PART 0D · SUBPART 78 · GUIDE 1048

Editorial Screening for AI-Assisted Research Fraud

What Should Journal Editors Look for in Potentially AI-Assisted Fraudulent Manuscripts?

Journal editors may encounter manuscripts containing fabricated data, invented methods, misleading figures, or manipulated citations, sometimes produced with generative AI. Learn which warning signs deserve attention, how to verify evidence, and how to distinguish legitimate AI-assisted scholarship from potentially fraudulent submissions.

PART 0D · SUBPART 78 · GUIDE 1047

AI Detection of Paper-Mill Patterns

Can AI Also Help Detect Paper-Mill Patterns?

AI can help detect recurring patterns across potentially fraudulent manuscripts, including duplicated images, unusual citation relationships, repeated methodological structures, and coordinated editorial activity. Learn which patterns are meaningful, how automated detection works, and why suspicious similarities require independent verification before any misconduct conclusion.