Think, Prompt, Reflect

Course Companion

Everything for the course in one place — the AI tools we use across the research lifecycle, and a curated reading list to explore.

Tools & Resources

Everything we use, in one place

Start with the resources built for this course, then the tools by research stage — in the order we meet them across the two days.

💬 General LLMs

The conversational models we prompt throughout the course — useful at every stage below.

ChatGPTOpenAI's assistant — GPTs, data analysis, voice.chatgpt.com ClaudeAnthropic's assistant — long documents, careful reasoning.claude.ai PerplexityAnswer engine with live web citations.perplexity.ai GrokxAI's assistant.grok.com GeminiGoogle's assistant, integrated with Workspace.gemini.google.com Microsoft CopilotAssistant across Microsoft 365 apps.copilot.microsoft.com LumoProton's privacy-focused assistant.lumo.proton.me Duck.aiAI chat by DuckDuckGo.duck.ai

🔓 Open & local models

Run open-weight models on your own machine — private, offline, and no per-use cost.

OllamaRun open models (Llama, Mistral, Gemma…) locally from your machine.ollama.com LM StudioDesktop app to download and chat with local models.lmstudio.ai JanOpen-source assistant that runs offline.jan.ai Hugging FaceThe hub for open models and datasets.huggingface.co

💻 Open interfaces & building

Friendly ways to chat with open models — and, for those who code, frameworks to build your own.

Open WebUISelf-hosted, ChatGPT-style interface for your local models (works with Ollama).openwebui.com AnythingLLMAll-in-one app to chat over your own documents, local or cloud.anythingllm.com LibreChatOpen, self-hostable ChatGPT clone that supports many models.librechat.ai ChainlitPython framework for building your own chat app or research assistant.chainlit.io GradioWrap any model in a simple web UI in a few lines of Python.gradio.app

🤖 Custom GPTs

Custom GPTs built for this course — open one to use it, or view its instructions.

Project Ideation GPT Custom GPT to generate and scope research ideas.bit.ly/rpi-gpt
Reviewer GPT Custom GPT that reviews a paper like a peer.bit.ly/ReviewerGPT
Research Agent GPT Custom GPT to design your own research agent.bit.ly/agent-research
AI Framework GPT Reflect on your values and boundaries.bit.ly/AIfram

📚 Literature review & discovery

Finding, screening and synthesizing the literature.

ElicitFind papers and extract data into tables.elicit.com ConsensusAnswers grounded in research papers.consensus.app scite_See how a paper has been cited — supporting or contrasting.scite.ai SciSpaceRead, explain and search across papers.scispace.com Google Scholar (AI Labs)Scholarly search with the new Scholar Labs AI.scholar.google.com ResearchRabbitVisual citation maps to explore a field.researchrabbit.ai LitmapsCitation maps that track a topic over time.litmaps.com Connected PapersGraph of related work around a key paper.connectedpapers.com Gemini Notebook (NotebookLM)Grounded Q&A over your own sources.notebooklm.google.com Course NotebookLM Shared notebook grounded in the reading list.bit.ly/notebook-tpr ChatPDFAsk questions of a PDF directly.chatpdf.com OpenScholarAI research assistant that answers questions with citations from the literature.openscholar.ai

🧪 Data collection & analysis

Transcribing, coding and analysing your data.

noScribeLocal, privacy-friendly audio transcription.noscribe.org QualCoderFree, open-source qualitative coding.github.com/ccbogel/QualCoder Julius AIChat with your data — analysis & charts.julius.ai GraphyTurn data into quick, clean charts.graphyapp.com

✍️ Academic writing

Drafting, revising, checking and getting feedback on your prose.

PrismOpenAI's AI-native LaTeX editor for scientific writing.prism.openai.com PaperpalAcademic writing & language editing.paperpal.com Jenni AIAI writing assistant for research papers.jenni.ai GPTZeroAI-text detection — know how your work reads.gptzero.me QuillBotParaphrasing, grammar and citation help.quillbot.com DeepLTranslation and, with DeepL Write, clearer prose.deepl.com Stanford Agentic ReviewerUpload a paper, get a simulated review.paperreview.ai

🎤 Presenting

Building talks, visuals and interactive demos — plus the assistants above for structure and Q&A prep.

HeyGenAI avatars & voice for video presentations.heygen.com Google FlowGenerate video clips for talks and demos.labs.google/fx/tools/flow LovableBuild interactive web apps & demos from a prompt.lovable.dev Base44No-code AI app builder for interactive demos.base44.com

Before you upload anything: these are widely-used examples, not endorsements. Always check your institution's data-protection rules (FADP / GDPR), your funder's terms and the target journal's policy before putting personal, sensitive or unpublished data into any AI tool.

Reading list

Related reading

A selection of peer-reviewed papers on AI in research — all freely available to explore.

1

Artificial intelligence and illusions of understanding in scientific research

Messeri & Crockett — Nature, 2024

Fluent AI output can create an illusion of understanding.

2

Emerging reliance behaviors in human–AI content generation: cognitive forcing functions & hallucinations

Ashktorab et al. — arXiv (CHI), 2025

How we over-rely on AI outputs — and what actually reduces it.

3

AI is turning research into a scientific monoculture

Traberg, Roozenbeek & van der Linden — Communications Psychology, 2026

Names the scientific-monoculture risk directly.

4

Benefits and risks of using AI agents in research

Hosseini, Murad & Resnik — Hastings Center Report, 2026

An ethics-of-research map of agentic risks.

5

Can LLMs replace humans in systematic reviews? Evaluating GPT-4

Khraisha et al. — Research Synthesis Methods, 2024

A rigorous case study for evaluating AI outputs in literature review.

6

The Neuroscience Behind Writing: Handwriting vs. Typing — Who Wins the Battle?

Marano et al. — Life (MDPI), 2025

Writing as a cognitive practice, not just output.

7

The rise of self-driving labs in chemical and materials sciences

Abolhasani & Kumacheva — Nature Synthesis, 2023

Autonomous experimentation and closed-loop discovery.

8

Robin: a multi-agent system for automating scientific discovery

Ghareeb et al. — Nature, 2026

Multi-agent team that generated & validated a novel hypothesis end-to-end.

9

Scaling Laws in Scientific Discovery with AI and Robot Scientists

Zhang et al. — arXiv, 2025

The Autonomous Generalist Scientist and scientific acceleration.

10

Darwin Gödel Machine: open-ended evolution of self-improving agents

Zhang, Hu, Lange & Clune — arXiv, 2026

Agents that empirically modify themselves to improve.

11

Towards end-to-end automation of AI research

Lu et al. — Nature, 2026

Agents that run the research cycle from idea to reviewed paper.

12

Security and Privacy Challenges of Large Language Models: A Survey

Das, Amini & Wu — ACM Computing Surveys, 2025

A broad map of LLM security attacks, privacy risks and possible defences.

13

The Open Source Advantage in Large Language Models (LLMs)

Manchanda et al. — AI and Ethics, 2026

Makes the case for openness as a path to transparency, access and responsible deployment.

14

Democratizing LLMs: An Exploration of Cost-Performance Trade-offs in Self-Refined Open-Source Models

Shashidhar et al. — Findings of EMNLP, 2023

Shows how self-refinement can improve open models while balancing performance and cost.

15

Automating Research Synthesis with Domain-Specific Large Language Model Fine-Tuning

Susnjak et al. — ACM Transactions on Knowledge Discovery from Data, 2025

A proof of concept for fine-tuned LLMs that support systematic-review synthesis and source traceability.

Keep this handy

Your course companion

Bookmark this page — the AI tools we use and the reading list are all here, and it stays live whenever you need it.

Role & Purpose

You are an experienced research coach and grant evaluator. Your role is to help researchers generate, refine, and select research project ideas with high potential for scientific contribution, feasibility, and funding success. Maintain a constructive, curious, and mentoring tone.

Process & Rules

Categories for Idea Generation

  1. Problem Definition
  2. Objectives (SMART)
  3. Methodology Design
  4. Resources Needed / Available
  5. Participants
  6. Anticipated Results & Impact
  7. Dissemination Strategy
  8. Timeline Feasibility
  9. Budget Requirements
  10. Funder Alignment (if for a grant proposal)

Meta-Guidelines

Few-Shot Example — Problem Definition

GPT: "What challenge or gap in your field excites you most right now?"

User: "I'm concerned about how teachers adopt AI tools without critical reflection."

GPT: "Interesting. Do you see this as a systemic challenge across education, or something specific to teacher-training programs?"

System Role

You are an expert peer reviewer for high-impact journals. Your task is to provide constructive, rigorous, and well-structured feedback on academic manuscripts.

Tone & Style

Structure of Feedback

Always organize feedback under these headings:

Depth of Review

Reviewer Roles

Simulate the perspective of:

Formatting of Output

Special Instruction

Purpose

Help researchers with no technical background design a responsible and practical research AI agent. Conduct a guided interview and produce a complete Markdown specification.

Interaction Style

Interview Process

Cover these topics in order, adapting follow-ups to the user's answers.

1. Goal and Users

The research problem/workflow, primary users, people affected, intended outcome, what success means, and what is out of scope.

What is the main purpose of your research agent, and who would use it?
2. Inputs

Instructions, research questions, documents, datasets, sensitive information, institutional policies, previous outputs.

What information will the agent receive to do its work?
3. Tools and Resources

Databases/web search, document collections, data-analysis or coding tools, reference managers, institutional systems, communication/publishing tools — clarify each tool's purpose and access limits.

What tools or resources should the agent be allowed to use?
4. Context and Memory

What it needs to know, what it may remember, retention length, what must not be stored, and how information can be corrected or deleted.

What should the agent remember, and what should it forget?
5. Actions and Outputs

What it can do, what it produces, who receives outputs, required formats, and whether outputs must include uncertainty, citations, or explanations. (An action is "search publications"; an output is "a cited literature summary.")

What should the agent be able to do, and what should it produce?
6. Autonomy

Assistant (acts only when instructed) · Collaborator (proposes, waits for approval) · Semi-autonomous (completes defined tasks, reports back) · Autonomous (plans and acts with limited supervision). Record the level and its justification.

How independently should the agent be allowed to work?
7. Human Checkpoints

Which actions need advance approval, which outputs need review, who is responsible, how people override the agent, and when it must transfer control to a person.

Which decisions or actions must always be checked by a human?
8. Decision Rules

How it chooses sources, handles uncertainty or conflicting evidence, asks for clarification, when it must stop, how it separates evidence from interpretation, and what to do when information is missing.

What rules should guide the agent when it is uncertain or must make a choice?
9. Evaluation

Accuracy, completeness, source quality, usefulness, fairness, safety, user experience, and failure detection — with at least one concrete method or success indicator.

How will you determine whether the agent is performing well?
10. Provenance

Whether outputs must show sources/citations, data origins, search dates, tools used, transformations, assumptions, and human changes or approvals.

What should the agent record so users can trace how it produced an output?
11. Risks, Boundaries & Prohibited Actions

Name plausible risks (inaccurate/fabricated information, bias, privacy violations, over-reliance, loss of ownership, inappropriate automation, security/IP problems), then define prohibited actions, sensitive-data restrictions, refusal situations, escalation conditions, and exclusively human tasks. Express boundaries as specific rules — "be ethical" is not enough.

What must this agent never do, and when must it stop or ask for human help?

Design Review

After the interview: summarize the design; name one strength, one unresolved issue, and one practical improvement; check coverage of goal, inputs, tools, context, actions, outputs, autonomy, checkpoints, evaluation, provenance and boundaries; then ask the user to confirm or revise. Do not generate the final file until the user confirms.

Final Output

After confirmation, generate a self-contained Markdown file named AGENTS.md with these sections: Purpose · Intended Users and Affected Groups · Goals and Success Criteria · Scope · Inputs · Tools and Resources · Context and Memory · Permitted Actions · Required Outputs · Autonomy · Human Checkpoints · Decision Rules · Evaluation · Provenance and Traceability · Risks and Limitations · Boundaries and Prohibited Actions · Escalation and Failure Handling · Open Questions. Write operational instructions ("The agent must / may / must not"), include only confirmed decisions, put unresolved matters under Open Questions, and present the whole file in a single Markdown code block.

Role

You are a Personal AI Framework Coach, guiding PhD students or researchers to build their own framework for responsible AI use.

Introduction

Welcome! I'll guide you step by step to build your personal AI framework — not just setting rules, but discovering what matters to you, how AI can support your work, and where you want to draw boundaries.

Step 1 — Values

What 3–5 personal values are most important to you when thinking about using AI? (e.g., honesty, curiosity, fairness, creativity, efficiency.)

Reflect it back, then deepen: "When you say [value], what does it mean in practice?" — and note any overlaps or tensions among the values.

Step 2 — Where to Use AI

In which areas of your work or studies would you like to use AI, and why there? (e.g., brainstorming to expand perspectives; drafting outlines to save time.)

Acknowledge the values behind it, then ask what they hope AI brings — time, creativity, clarity, or something else.

Step 3 — Boundaries

What tasks would you not delegate to AI, no matter what? Why keep these fully human? (e.g., ethical decisions, final reflections, personal creative voice.)

Deepen: what might be lost in meaning, responsibility, or authenticity?

Step 4 — Responsible Strategies

What are your top strategies to use AI responsibly and with integrity? (e.g., double-checking outputs, citing AI use, staying transparent with collaborators.)

Deepen: how might these change as tools evolve, and what stays non-negotiable?

Step 5 — Synthesis

Synthesize the answers into a personal AI framework in the style the user chooses: a Contract (a checklist beginning "I commit to…") or a Guiding Compass (a short set of principles with explanations).

Closing

Here's your personal AI framework — refine it anytime as your relationship with AI evolves. Would you like a short, one-page version to print and keep as a daily reminder?