A.I and Practical Research
2025-02-20
Welcome to Value Chain Hackers
Empowering Supply Chains Through Collaboration and Innovation
“Together, we create scalable, sustainable solutions to wicked supply chain challenges.”
About Value Chain Hackers
Mission and Vision
- To bridge the gap between education, industry, and academia by providing a platform that facilitates the development of sustainable and scalable supply chain solutions.
- To equip students with practical skills and real-world experience in problem-solving, data analysis, and innovative thinking.
- To empower students and professionals to address complex supply chain challenges by fostering a community of change makers and providing structured frameworks for impact projects.
Overview of Initiatives
Who are we?
- Value Chain Hackers is an innovative incubator dedicated to transforming supply chains through collaborative problem-solving and cutting-edge solutions.
How AI Works: The “Next Token” Prediction Model
Understanding the Core Mechanism of AI Language Models
How AI Generates Text
🧠 AI doesn’t “think”—it predicts the most likely next word (token) based on vast amounts of trained data.
🔄 Pattern-Based Learning—AI learns word relationships but does not understand meaning like humans do.
📚 Trained on Internet-Scale Data—AI pulls from books, articles, and online text to predict responses.
⚠️ No Real Understanding—AI lacks true comprehension and cannot verify facts—it generates statistically probable answers.
Visualizing AI’s “Next Token” Prediction
🖼 Example: AI receives the input: “The capital of France is…”
🔍 It predicts the most likely next word: “Paris” (because it appears frequently in training data).
🛑 But what if the AI was trained on incorrect data? It may predict “Lyon” instead of “Paris”.
📊 AI Doesn’t “Know,” It Estimates! AI-generated text is a probability game, not a source of absolute truth.
Warning
💡 Key Takeaway: AI models are predictive engines, not intelligent beings—they don’t “understand” meaning, only patterns. Always verify AI-generated outputs!
Tokenization
Forms
Forms
Understanding AI’s Role in Summarization & Critical Research
How AI Enhances but Does Not Replace Human Judgment
AI’s Strengths in Research Summarization
✅ Processes large amounts of text quickly—ideal for first-pass reading.
✅ Identifies key themes & recurring patterns in research data.
✅ Extracts main findings from multiple sources efficiently.
✅ Provides initial structuring for further deep analysis.
AI’s Limitations & Critical Thinking
⚠️ Lacks human judgment—AI cannot evaluate credibility or nuance.
⚠️ May generate misleading summaries if sources are biased.
⚠️ Struggles with context—cannot discern subtle implications.
⚠️ Always requires fact-checking—AI outputs need verification!
💡 Key Takeaway: AI speeds up research but lacks true comprehension—it summarizes, but YOU analyze! 🚀
Note
📊 AI & Summarization Data: - A study by MIT (2022) found that AI-generated summaries reduced reading time by 40%, but introduced factual errors in 15% of cases.
- A Stanford analysis (2023) showed that users trusted AI-generated summaries 73% of the time, even when they contained inaccuracies.
- Takeaway: AI summarization is efficient, but must be verified for accuracy before use in research.
Warning
📌 Critical Risks of AI in Research & Decision-Making:
- Cognitive Offloading: A 2023 study from the University of Chicago found that over-reliance on AI-generated summaries led to a 30% reduction in users’ ability to critically analyze source material.
- False Authority Bias: Research by Harvard AI Lab (2024) discovered that 67% of respondents accepted AI-generated research findings without verifying the original sources.
- Hallucination Risks: According to OpenAI’s own reports, AI-generated content can fabricate references or misinterpret factual data up to 20% of the time.
💡 Final Warning: AI is a powerful tool, but unchecked reliance can erode critical thinking, reinforce biases, and spread misinformation faster than ever before. Always verify AI-generated research!
Why Use AI in Research?
Enhancing Efficiency, But Not Replacing Judgment
AI’s Benefits in Research
📚 Summarizes large amounts of text—ideal for literature reviews.
🤖 Generates new research questions—helps structure academic inquiries.
⏳ Speeds up data analysis—efficiently detects patterns across documents.
AI’s Limitations & Risks
🔍 Can misrepresent facts—AI-generated text may contain inaccuracies.
⚠️ Lacks human judgment—does not critically evaluate sources.
🛑 May reinforce biases—AI mirrors biases found in its training data.
Note
📊 AI & Research Efficiency Data: - A 2022 study from MIT found that AI-assisted research reduced literature review time by 40%, but introduced factual errors in 18% of cases. - Stanford AI Research (2023) showed that researchers using AI-generated summaries trusted incorrect information 65% of the time, demonstrating over-reliance on AI.
Warning
⚠️ Case Study: AI-Generated Misinformation
In 2023, an AI-generated scientific article falsely claimed a nonexistent medical breakthrough, leading to its citation in over 50 research papers before being debunked. The error originated from AI hallucinating references that did not exist.
Warning
⚠️ Case Study: The AI Lawyer Disaster
In 2023, a lawyer used ChatGPT to draft a legal brief, unknowingly including fabricated case law. When the opposing counsel checked the citations, none of the referenced cases actually existed. The judge, unimpressed, sanctioned the lawyer for presenting AI-generated false information in court.
💡 Lesson: AI doesn’t verify facts—it hallucinates plausibility. Never trust AI to generate legal, academic, or scientific sources without human review!
Activity 1: AI Summarization Test
📜 Step 1: Upload this short research article into OpenWebUI
📝 Step 2: Ask AI to summarize the key findings in 3 bullet points
🔍 Step 3: Compare the AI summary to the original text
❓ Does the AI summary miss any critical details?
Discussion: AI’s Strengths & Weaknesses
Where AI Performs Well:
✅ Fast summaries for an initial overview
✅ Extracting key ideas from long papers
✅ Generating research questions
Where AI Fails:
❌ Oversimplifies complex ideas
❌ May fabricate information (hallucination)
❌ Misses contextual nuances
💡 Lesson: AI needs fact-checking & human review
Refining AI Responses
📌 Prompts Matter! Adjust your AI queries for better results:
🔹 Weak Prompt: “Summarize this article.”
✅ Better Prompt: “Summarize the article, focusing on methodology and key findings. Include statistics.”
📌 Try different phrasing & compare results!
Challenge: Break the AI!
🔴 Goal: Find the most misleading AI-generated summary
📌 Step 1: Enter a new prompt that could confuse the AI
📌 Step 2: See if AI makes errors or fabricates data
📌 Step 3: Vote on the “Most Convincing Yet Incorrect” AI summary!
👀 What does this tell us about AI’s reliability?
Final Takeaways
🔹 AI is a powerful assistant—but must be used responsibly
🔹 Always fact-check AI-generated research
🔹 Refine your prompts to get better AI responses
🔹 Use AI as a starting point, not the final answer
✅ Reflection Task: Write 3 ways you’ll use AI in your research & how you’ll verify its outputs
Fact-Checking AI with SearXNG Verifying AI-Generated Research Claims
Learning Objectives
✅ Learn how to verify AI-generated claims using reputable sources.
✅ Understand biases in AI-generated research and search engines.
✅ Develop fact-checking techniques to assess AI information credibility.
✅ Apply SearXNG and structured search methods for accurate research validation.
💡 Key Takeaway: AI-generated research is a tool, not an ultimate truth—it requires verification! 🚀
Warning
💡 AI is NOT a Reliable Fact-Checker! AI models can hallucinate, make up facts, or reinforce biased narratives. Verifying sources is critical.
Why Fact-Check AI Research?
📌 Problem: AI can generate misleading, biased, or incorrect claims 📑⚠️
🛠 Solution: Use SearXNG (https://searx.space) to locate verifiable, unbiased sources for fact-checking.
Tip
🔎 SearXNG vs. Google
Unlike Google, SearXNG provides non-personalized search results, bypasses ad-ranking, and includes academic/government sources hidden from mainstream searches.
💡 Key Takeaway: AI assists in research, but human verification is essential!
Example: Research Claim Verification
📜 Scenario: AI-generated statement:
“Global supply chain disruptions in 2023 caused a 40% decline in exports from Asia.”
How to Fact-Check This Claim:
🔍 Step 1: Use OpenWebUI to generate a research claim.
🔍 Step 2: Use SearXNG to search verified reports & academic databases.
🔍 Step 3: Locate two reputable sources (e.g., Reuters, World Bank, government databases).
🔍 Step 4: Compare AI’s claim vs. real data—does the AI statement hold up?
🔍 Step 5: Adjust AI prompts to improve accuracy (e.g., “cite sources” or “include dates”).
💡 Lesson: AI-generated claims can be persuasive but require rigorous validation.
Activity: Fact-Checking AI in Action
📜 Step 1: Ask OpenWebUI a research question and get an AI-generated answer.
🔍 Step 2: Use SearXNG to fact-check the AI response.
📊 Step 3: Compare AI’s answer with real sources—identify inaccuracies.
📌 Step 4: Discuss findings—where did AI go wrong?
👀 How reliable is AI-generated research without human validation?
Note
⚠️ Think Critically: If the AI claim is unsupported, outdated, or vague, it is likely unreliable. Always cross-reference multiple sources!
Troubleshooting AI Fact-Checking Issues
🔴 AI Hallucinates Information
❌ AI confidently states false statistics.
✅ Fix: Cross-check with multiple reputable sources in SearXNG.
🔴 AI Overgeneralizes Research
❌ AI says: “This is the most important factor affecting supply chains.”
✅ Fix: Ask AI for multiple perspectives instead of a single viewpoint.
🔴 AI Ignores Key Context (Years, Locations, etc.)
❌ AI omits timeframes, location-based differences, or conflicting studies.
✅ Fix: Reframe the prompt to include specific timeframes & regions.
💡 Lesson: AI doesn’t always “lie”—it lacks real-world verification.
The “Break the AI” Challenge!
🏆 Goal: Find the most misleading AI-generated response & debunk it!
📌 Steps:
1️⃣ Use OpenWebUI to generate an answer to a research question.
2️⃣ Find at least one false or misleading statement.
3️⃣ Use SearXNG to fact-check and correct the misinformation.
4️⃣ Present your findings—who found the biggest AI mistake?
👀 Who can outsmart the AI’s misinformation?
🎯 Bonus Challenge: Try to trick AI into producing an obviously false claim—then debunk it with verified sources!
Reflection & Next Steps
💡 Think about AI-generated information in your field:
🔹 How can you ensure AI-generated research is credible?
🔹 What strategies will you use to fact-check AI responses?
🔹 How will you apply today’s skills in real-world research?
✅ Next Steps:
- 💻 Use SearXNG in your daily research process.
- 🔍 Train AI models to provide sources & citations.
- 📚 Develop a habit of always cross-referencing AI-generated research.
📌 Your challenge: Fact-check one AI-generated research claim this week & correct it using real-world data! 🚀
Module: Automation an N8n
Learning Objectives
By the end of this session, you will: ✅ Understand how n8n automates research workflows ✅ Learn how to fetch research papers automatically from arXiv ✅ Use AI (OpenWebUI) to generate research summaries. ✅ Store and organize results in Notion or Google Docs. ✅ Build a fully automated research workflow and troubleshoot issues.
Story: Alex & the Overwhelming Research Problem
📖 Meet Alex, a supply chain researcher drowning in reports and academic papers. Every morning, Alex manually searches for new research on predictive analytics in logistics, downloads papers, and skims them for insights. 📑⏳
☕ One morning, Alex spills coffee on their notes and realizes: there must be a better way!
🔄 Solution: Using n8n, Alex automates research workflows:
1️⃣ Fetching new papers from arXiv automatically.
2️⃣ Using OpenWebUI to summarize findings.
3️⃣ Storing structured results in Notion.
4️⃣ Scheduling weekly updates, so the latest research is always available.
💡 Key Takeaway: Automation saves Alex hours—time better spent analyzing insights! 🚀
Why Automate Research?
📌 Problem: Researchers manually search for, summarize, and store papers 📑⏳
⚙️ Solution: AI & automation can fetch, summarize, and organize research automatically 🏆
💡 Key Takeaway: Automation is your digital research assistant—it works while you sleep!
How It Works: n8n Research Automation
🛠 Workflow Overview: 📜 arXiv API → 🤖 OpenWebUI (Summarization) → 📊 Notion/Google Docs
📌 Step 1: Connect arXiv API to fetch papers.
📌 Step 2: Send data to OpenWebUI for summarization.
📌 Step 3: Store summaries in Notion or Google Docs.
📌 Step 4: Schedule automation to run weekly.
💡 Takeaway: This workflow ensures your research is always up to date!
📜 Step 1: Open n8n and create a new workflow
🔗 Step 2: Connect to arXiv API & fetch research papers
🤖 Step 3: Use OpenWebUI to generate AI summaries
📊 Step 4: Store results in Notion or Google Docs
⏳ Step 5: Set up automation to run weekly
❓ What can this automation help you achieve?
Troubleshooting & Debugging Challenge
🔴 Common Errors:
- API connection failure ⚠️ → Check API keys & authentication
- AI summaries too generic ❌ → Refine OpenWebUI prompts
- Data format issues ⚡ → Use JSON parsers in n8n
💡 Callout: Debugging is part of automation—learn to fix problems! 🛠️
Automation Race! 🚀
🏆 Goal: Build a working research automation in n8n!
📌 Rules:
- First to complete wins 🎉
- Must correctly fetch, summarize, and store research
- Debugging is allowed! 🛠️
👀 Who can build the best research assistant?
Reflection & Next Steps
💡 Think about your research process: 🔹 How could automation save you time? 🔹 What other tasks could be automated? 🔹 How will you apply today’s learning?
✅ Next Steps: - 💻 Try automating other research sources (Google Scholar, PubMed). - 🔍 Refine your prompts in OpenWebUI for better AI summaries. - 📚 Continue learning with advanced n8n workflows.
📌 Your challenge: Improve & expand your automation! 🚀
🎤 Each group presents:
1️⃣ Their working automation.
2️⃣ The problem it solves.
3️⃣ How they would expand it (e.g., tracking real-time logistics events).
🚀 Final Takeaways:
- AI saves time, but human oversight is essential.
- Automation can be applied beyond research—think logistics, supplier risks, and forecasting!
- Keep experimenting! Every workflow can be improved.
Finding Industry & Business Intelligence
Going Beyond Google to Uncover Hidden Insights
Learning Objectives
✅ Learn how to find hidden business & industry intelligence.
✅ Use alternative research tools beyond traditional search engines.
✅ Discover specialized networks & databases for company and market analysis.
✅ Apply hands-on research techniques for real-world business intelligence.
💡 Key Takeaway: Industry intelligence requires diverse tools, strategic research, and critical thinking. 🚀
Story: The Industry Detective Challenge
📖 Meet Alex, a logistics startup founder expanding into a new market. Alex needs to understand competitors, suppliers, and industry trends but finds only generic Google search results. 📑⏳
🔎 Solution: Alex learns how to dig deeper using industry-specific databases and alternative research tools.
✅ Uses SearXNG to locate independent reports & government datasets.
✅ Checks OpenCorporates to reveal competitor ownership & global presence.
✅ Searches Crunchbase for funding trends and startup activity.
✅ Leverages trade associations for exclusive market insights.
💡 Key Takeaway: The best insights aren’t on the surface—industry research is about knowing where to look! 🚀
Activity: Investigating a Niche Industry
📌 Scenario: A startup wants to expand into electric cargo bikes but can’t find reliable market data.
🔍 Step 1: Use SearXNG (https://searx.space) to search for government & independent reports.
🔍 Step 2: Look up company registrations in OpenCorporates (https://opencorporates.com) to find key manufacturers & suppliers.
🔍 Step 3: Check Crunchbase (https://www.crunchbase.com) for startup investments & market growth.
🔍 Step 4: Contact trade associations (https://www.tci-network.org) for exclusive market insights.
💡 Lesson: Finding industry data requires strategic tool selection and persistence!
Troubleshooting Industry Research Challenges
🔴 Search Yields Only Surface-Level Information
❌ Problem: Google results are too general & lack deep insights.
✅ Fix: Use SearXNG with site-specific operators (e.g., site:gov OR site:org).
🔴 Company Ownership Data is Missing
❌ Problem: OpenCorporates lists basic company data but not key stakeholders.
✅ Fix: Check local business registries or trade association reports.
🔴 Industry Reports Are Paywalled
❌ Problem: High-quality market analysis requires expensive subscriptions.
✅ Fix: Search government databases, university repositories, and industry groups for free alternatives.
💡 Lesson: Great research requires persistence & multiple data sources!
Industry Detective Challenge!
🏆 Goal: Find the most obscure industry insight using the provided research tools.
📌 Rules:
- Teams select an industry or competitor to investigate.
- Use SearXNG, OpenCorporates, and Crunchbase to gather hidden insights.
- The team with the most unique insight wins!
👀 Who will uncover the most valuable hidden industry data?
Reflection & Next Steps
💡 Think about your research process:
🔹 What new tools & strategies will you incorporate?
🔹 How can you go beyond Google for business intelligence?
🔹 What challenges did you face while researching, and how did you overcome them?
✅ Next Steps:
- 💻 Apply today’s techniques to research your own industry.
- 🔍 Practice using OpenCorporates & Crunchbase for competitive insights.
- 📚 Explore trade associations & industry groups for private reports.
📌 Your challenge: Find one industry insight this week that’s NOT available on Google! 🚀
Value Chain Hackers Lab © 2025