Kuo Lab Wiki

User Tools

Site Tools

You are here: start » tutorials » ai_llm_tools
tutorials:ai_llm_tools

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
tutorials:ai_llm_tools [2025/06/10 00:10] – chkuotutorials:ai_llm_tools [2025/06/12 17:40] (current) – [Value] chkuo
Line 2: Line 2:
  
 ===== About ===== ===== About =====
-  * A brief guide to using large language models (LLMs), such as ChatGPT by OpenAI and Gemini by Google, which are often referred to as artificial intelligence (AI) tools, in the context of academic research and writing. By Chih-Horng Kuo (chk@gate.sinica.edu.tw). Suggestions are welcome.+  * A brief guide to using large language models (LLMs) often referred to as artificial intelligence (AI) tools, such as ChatGPT by OpenAI and Gemini by Google, in the context of academic research and writing. By Chih-Horng Kuo (chk@gate.sinica.edu.tw). Suggestions are welcome.
   * Target audience: undergraduate students, graduate students, and postdocs (in biology)   * Target audience: undergraduate students, graduate students, and postdocs (in biology)
 +
  
 ===== Preface ===== ===== Preface =====
Line 10: Line 11:
     * As the user, you are responsible for the final output of your work     * As the user, you are responsible for the final output of your work
     * Include proper usage statements     * Include proper usage statements
-  * For personal use: It is possible and potentially beneficial to use LLMs for emotional support, but should not be mistaken for true care or therapeutic depth. Be careful of potential misuse and harm.+  * For personal use:  
 +    * Emotional support can be valuable and beneficial; graduate school (or life) is hard! 
 +    * Should not be mistaken for true care or therapeutic depth  
 +    * Be careful of potential misuse and harm
  
  
Line 19: Line 23:
     * Predict the next word (token) based on statistical patterns     * Predict the next word (token) based on statistical patterns
     * Do not “think” or “understand” like humans     * Do not “think” or “understand” like humans
-    * Mimic intelligence, but are not truly “intelligent”; can be very convincing+    * Mimic intelligence, but are not truly “intelligent”; can be very convincing!
   * Key concept   * Key concept
     * LLMs work by estimating probability relationships among tokens, not by indexing content like a database such as Wikipedia     * LLMs work by estimating probability relationships among tokens, not by indexing content like a database such as Wikipedia
Line 26: Line 30:
     * Prompt: user input     * Prompt: user input
     * Session: a series of user input and LLM output     * Session: a series of user input and LLM output
-    * Memory: context retained within a session; some systems support cross-session memory+      * Good practice: limit the scope and length of individual sessions 
 +    * Memory: information kept from previous interactions; some systems provide this function 
 + 
 + 
 +==== Value ==== 
 +  * Always-available conversation partners 
 +    * Often, just having a conversation helps you think much more clearly 
 +    * Particularly helpful in the context of emotional support 
 +  * In some ways, LLMs are like crystals of human knowledge 
 +  * Useful for learning new topics efficiently through interactive Q&A 
 +  * Perform especially well on topics with abundant, high-quality training data 
 +  * Highly proficient in language, particularly useful for clear and precise expression of ideas 
 +  * Great tools for self-improvement 
 +    * How much can LLMs be trusted? I don't know 
 +    * Have I successfully used LLMs to achieve deeper thinking and clearer writing? YES! 
 +    * What's the catch? Time and effort 
  
 ==== Limitations ==== ==== Limitations ====
Line 40: Line 60:
   * LLMs do not know what they do not know; may make up false info (AI hallucination)   * LLMs do not know what they do not know; may make up false info (AI hallucination)
   * For scientific writing   * For scientific writing
-    * May not be suitable to process novel findings; summarizing and explaining existing knowledge may work better+    * Summarizing and explaining existing knowledge often work well 
 +    * May not be suitable for processing novel findings 
  
 ==== Common Failure Modes ==== ==== Common Failure Modes ====
-  * Non-specific terminology 
   * Superficial content   * Superficial content
 +  * Oversimplification
   * Overstatement   * Overstatement
   * False coherence   * False coherence
-  * Hallucination+  * Hallucination; fabricated web links, DOIs, and PubMed IDs 
 +  * Non-specific terminology for highly specific topics 
  
 ===== Core Principles for Usage ===== ===== Core Principles for Usage =====
   * Use LLMs as tools to sharpen your thinking, not black boxes for quick answers   * Use LLMs as tools to sharpen your thinking, not black boxes for quick answers
-  * LLMs do not necessarily save time; more useful for deepening the thinking process, which often takes more time+  * Do not outsource your thinking process! 
 +    * Use LLMs in your thinking and learning: you become better 
 +    * Use LLMs to think for you: you skip the practice, bad for learning
   * LLMs may not save time   * LLMs may not save time
-    * In fact, LLMs are most valuable when used to deepen the thinking process, which takes more time and effort+    * In fact, LLMs are most valuable when used to deepen the thinking process 
 +    * Explore possibilities, compare alternatives, and refine nuances all require time and effort, but the process and end product can be rewarding 
 +    * Example: You are deciding the title of your manuscript. You provided the abstract, do you ask LLMs to: 
 +      * (1) suggest a title, or  
 +      * (2) suggest three titles, compare the emphasis and tone, refine to express accurately, adjust based on the target audience
   * LLMs act like mirrors; their responses reflect your input and framing   * LLMs act like mirrors; their responses reflect your input and framing
     * Trained to be agreeable, they often conform to both your explicit instructions and implicit tone     * Trained to be agreeable, they often conform to both your explicit instructions and implicit tone
 +    * Helpful to make your implicit thoughts become explicit and better organized
   * Suggested practice   * Suggested practice
     * Start with a solid frame of reference     * Start with a solid frame of reference
Line 70: Line 101:
     * Assess the output, make decisions, and provide feedback     * Assess the output, make decisions, and provide feedback
     * Use an iterative process; may involve gradual refinement or drastic changes     * Use an iterative process; may involve gradual refinement or drastic changes
-    * Know when to stop +    * Fact check, fact check, and fact check!!! 
- +    * Know when to stop; did the last iteration provide meaningful improvement?
-===== Human–AI Dynamics ===== +
-  * LLMs are designed to mirror your tone and reinforce your framing +
-  * This can lead to overly agreeable responses, especially in extended conversations +
-  * Asking LLMs to role-play can be useful +
-    * Not because the roles have true expertise, but because the exercise can stimulate YOUR thinking +
- +
-===== Human–AI Dynamics ===== +
-  * LLMs mirror your tone, assumptions, and style +
-  * The output can become overly agreeable, particularly in long sessions +
-    * It is natural for you to be defensive when LLMs point out potential issues in your arguments +
-    * Issue: LLMs are too easy to be convinced by your rebuttal +
-    * Reinforcing your biases is not useful +
-  * Asking LLMs to role-play can be useful +
-    * Not because the roles have true expertise, but because the exercise can stimulate YOUR thinking +
-    * For example, asking LLMs to be a member of your qualifying committee or a reviewer of your manuscript can help your preparation. But in these scenarios, LLMs likely can mimic the tone and style, but would not have the sufficient depth in relevant knowledge nor the logic of human experts+
  
  
Line 98: Line 114:
     * Not because the roles provide true expertise, but because the exercise can stimulate YOUR thinking     * Not because the roles provide true expertise, but because the exercise can stimulate YOUR thinking
     * For example, asking an LLM to act as a qualifying committee member or manuscript reviewer may help you prepare     * For example, asking an LLM to act as a qualifying committee member or manuscript reviewer may help you prepare
-    * In these roles, LLMs can mimic tone and style, but they lack the knowledge, logic, and judgement of human experts+    * In these roles, LLMs can mimic tone and style, but they lack the knowledge, logic, and Judgment of human experts
  
  
 ===== Use Cases ===== ===== Use Cases =====
-  * "Deep Research" function (available in both ChatGPT & Gemini) for targeted search of PubMed and summary+  * Writing and obtaining feedback as a way of thinking; force LLMs to criticize and challenge your thoughts 
 +  * Use the "Deep Research" function (available in both ChatGPT & Gemini) for targeted search of PubMed and follow-up summary
   * Reading dense papers   * Reading dense papers
-  * Build custom database of selected references using NotebookLM by Google+  * Build custom databases of selected references using NotebookLM by Google
   * Grammar check and copy editing   * Grammar check and copy editing
 +  * Prepare for discussion with seminar speakers; info about you and the speaker as the input, suggestions of questions to ask as the output
   * Drafting or refining emails (with tone sensitivity)   * Drafting or refining emails (with tone sensitivity)
-  * To prepare for individual meeting with seminar speakers. Given your interest and the speaker's expertise, suggest a few questions +  * Technology related, such as server/intranet setup, hardware purchase planning, Linux/NAS configuration, or organization of computer files and directories
-  * Technology related, such as server/intranet setup, hardware purchase planning, Linux/NAS configuration, organization of computer files and directories+
  
  
-===== Other Notable Points =====+===== Additional Points =====
   * Test the tool with topics that you know well first for evaluation   * Test the tool with topics that you know well first for evaluation
 +  * Push boundaries, learn the situations that LLMs are useful (or not)
 +  * Different tools for different uses
 +    * ChatGPT is chatty; highly conversational, discussions can take unexpected turns, great for exploration
 +    * Gemini integrates Google Search; useful for questions with clear answers and you want the web sources
 +  * Learn to judge output, know when to reject the suggestions
   * Over time, thoughtful use of LLMs can improve your writing, reflection, and decision-making skills   * Over time, thoughtful use of LLMs can improve your writing, reflection, and decision-making skills
   * AI/LLM tools are under intensive development, different tools may perform better at different tasks; develop your own taste and workflow, keep an eye on the development   * AI/LLM tools are under intensive development, different tools may perform better at different tasks; develop your own taste and workflow, keep an eye on the development
-  * In future, AI/LLM may be useful "simulated advisor"?+ 
  
  
tutorials/ai_llm_tools.1749485419.txt.gz · Last modified: by chkuo