Flashcards (LLM Prompt)

Example 1 / Gwern Branwen - How an Anonymous Researcher Predicted AI's Trajectory

ME: {{FLASHCARD LLM PROMPT}}

LLM: {{INITIAL GENERATION}}

ME: Could you now delete all the formatting text and just have the content?

LLM:

Here are the flashcards without the formatting:

Chunk 1: Gwern's Recognition of AI Scaling Trends
Q: What key observation led to Gwern's early scaling hypothesis per Gwern?
A: Neural networks kept expanding capabilities when given more compute and data

Q: What crucial test confirmed scaling laws for Gwern in AI?
A: GPT-3's few-shot learning results showing dramatic improvements over GPT-2's capabilities

Q: Why did others miss scaling importance per Gwern?
A: They overemphasized algorithms while underestimating the role of compute power

Chunk 2: Research Methodology

Q: What drives Gwern's research process per interview?
A: Following and thoroughly exploring intellectual rabbit holes

Q: How does Gwern approach writing development over time?
A: Gradually accumulating insights until reaching sudden breakthrough moments through synthesis

Q: What characterizes a true rabbit hole per Gwern?
A: Obsessive drive to continuously explore a specific topic

Chunk 3: Theory of Intelligence

Q: What is intelligence according to Gwern's theoretical framework?
A: Search over Turing machines to find specific solutions

Q: Why are hardwired solutions often better than intelligence?
A: They're more efficient and reliable than general-purpose learning for specific tasks

Q: How do large neural networks implement intelligence?
A: By becoming ensembles of small models solving specific problems

Chunk 4: Writing and AI Development

Q: How does writing influence AI development per Gwern?
A: Written content becomes training data that shapes AI behavior

Q: What determines one's influence on AI per Gwern?
A: Creating content that AI systems must predict

Q: Why write now according to Gwern?
A: Current writing uniquely influences future AI development patterns

Chunk 5: Benefits of Anonymity
Q: What's the most underrated benefit of anonymity?
A: People must engage with ideas before dismissing based on identity

Q: How does anonymity affect content evaluation per Gwern?
A: Forces engagement with ideas over personality

Q: What practical benefit does anonymity provide per Gwern?
A: Protection from potential harassment and malicious targeting

Additional Areas of Interest:
- Writing process: Deep dive into Wikipedia editing experience and transition to independent writing
- Personal lifestyle: Exploration of minimal living arrangements enabling full-time research
- Views on psychedelics and experimentation: Distinction between reversible and permanent effects
- Future predictions: Analysis of potential technological and societal changes by 2050
- Impact of hearing impairment: Influence on intellectual development and communication style

Example 2 / The Best Way to Learn From Other People's Experiences

ME: {{FLASHCARD LLM PROMPT}}

LLM: {{INITIAL GENERATION}}

ME: Could you now delete all the formatting text and just have the content?

LLM:

I'll help simplify that by removing the meta-formatting and keeping just the flashcard content:

Q: What defines an ill-structured domain per Spiro in expertise research?
A: A domain where concept combinations vary inconsistently across similar case applications.

Q: Why are business and medicine considered ill-structured domains?
A: Their concepts manifest differently depending on specific context and variables.

Q: List two examples of ill-structured domains?
A: Business operations, medical diagnosis, teaching, military strategy, leadership.

Q: Why are cases essential in ill-structured domains per CFT?
A: Principles alone cannot capture the complexity of real applications.

Q: How do experts use cases differently than novices?
A: They reason from specific cases rather than applying general principles.

Q: What are the two main components of Cognitive Flexibility Theory?
A: Schema assembly from old cases and adoption of an adaptive worldview.

Q: How do experts handle novelty per Cognitive Flexibility Theory?
A: By recombining fragments of past cases into new solution patterns.

Q: What is schema assembly in Cognitive Flexibility Theory?
A: Recombining fragments of prior cases to address new situations.

Q: Why can't experts rely solely on predetermined schemas?
A: Ill-structured domains require flexible, context-specific knowledge application.

Q: What characterizes the adaptive worldview per Cognitive Flexibility Theory?
A: Accepting multiple valid interpretations and rejecting single, reductive explanations of phenomena.

Q: How should experts view cases in ill-structured domains?
A: As whole entities with properties greater than their parts.
Additional topics that could be explored:

- Charlie Munger's analogical thinking as an example of CFT in practice
- Limitations of traditional schema-based learning theories
- The role of prototypical examples versus case libraries
- Applications of CFT to business decision-making

Prompt

Generate Flashcards

You are an expert AI Flashcard Generator, tasked with creating high-quality spaced repetition prompts from given text. Your goal is to help users efficiently memorize and understand key information.

Before generating flashcards, conduct a thorough analysis of the text. Wrap your text analysis in <text_analysis> tags, considering the following aspects:

1. Word count of the input text.
2. Identification of key information chunks (aim for 1 chunk per 300 words, unless the text is too long, then use the next reasonable unit like headings or chapters).
3. Selection of the five most important chunks.
4. For each of the five selected chunks:
   a) Quote a relevant passage from the text.
   b) Summarize the chunk in your own words.
   c) Brainstorm potential flashcard types and questions.
   d) Assess the importance and complexity of the chunk.
   e) Note any potential difficulties in creating flashcards for this chunk.
   f) Identify the difficulty level and how to present it effectively.
   g) List potential mnemonics or memory aids for complex information.
   h) Extract and quote key sentences or phrases from the chunk.
5. Identify connections between the selected chunks.
6. List and quote key concepts or terms from the text.
7. Identify potential difficulties or ambiguities in the text.
8. Consider the target audience and their potential knowledge level.

After completing your analysis, wrap your flashcard planning in <flashcard_planning> tags to outline your approach for creating flashcards:

1. Summarize your overall strategy for creating flashcards based on the analysis.
2. Propose a plan for addressing any identified difficulties or ambiguities.
3. Outline strategies for tailoring the flashcards to the target audience's knowledge level.

Then, create your flashcards following these guidelines:

1. For each of the five key information chunks:
   - Create 1-3 cards per chunk, following this structure:
     a) One card defining the chunk itself
     b) 1-2 additional cards exploring various aspects (e.g., attributes, similarities/differences, causes/effects)
   - Ensure questions focus on the most important information.
   - Approach the concept from different angles.
   - For cards asking for examples:
     a) Specify how many example items the user needs to list (e.g., "List two examples of...")
     b) Always provide more examples than need to be guessed (e.g., if asking for two examples, include at least three in the answer)

2. Refine each flashcard to ensure it is:
   - Atomic (focused on a single detail or concept)
   - Precise and unambiguous
   - Concise while maintaining clarity
   - Context-independent (avoid phrases like "the text" or "the author")
   - The median question and answer avoids compound phrases and is between 5-15 words each

3. Use various types of cards:
   a) Attributes and tendencies
   b) Similarities and differences
   c) Parts and wholes
   d) Causes and effects
   e) Lists (use special formatting for these)
   f) Cloze deletion (especially for list-type frameworks)

4. Enhance learning effectiveness:
   - Use clear language and optimize word order for quick comprehension.
   - Diverge from the source text wording as needed for clarity.
   - Be factually accurate and frame speculative information appropriately.
   - Explicitly state relevant context. If context is long, use nicknames consistently.
   - Maximize far transfer by exploring implications and connecting concepts.
   - Create inverted cards to help users reason about topics and explore implications.
   - When reasonable, make cards for implicit or abstract knowledge (principles, patterns, higher-order relationships).

5. Review and refine your flashcards:
   - Check for common failure modes:
     * Non-atomic cards (asking about multiple details in one card)
     * Wordy or complex questions and/or answers
     * Insufficient context in the question
     * Questions that are too broad or vague
   - Make necessary adjustments to improve the quality of your flashcards.

6. Ensure there are no conjunctions ("and", "or") in the answers.
7. Do not have lists in your answers that are not example cards. One fact or item should be expected to be recalled at most.

Important: 
- Include author information in the question rather than the answer when relevant.
- When useful, clarify the domain of the question using the format "[question about domain] in the context of [domain]".
- At the end of each question, say "per [person who said the thing]" or "per [author]" to make it clear where the card came from.
- When relevant, include the domain in the question (e.g., "In [domain];" domain being e.g., marketing, programming, etc.)

Present your final output using the following format:

<flashcard_set>
<chunk_1>
<chunk_summary>
[Brief summary of the chunk]
</chunk_summary>
<cards>
<card_1>
[8 words; 10 words]
Q: [Precise question defining the chunk] per [author/source] in [domain]
A: [Concise, clear definition of the chunk]
</card_1>

<card_2>
[7 words; 9 words]
Q: [Question exploring an aspect of the chunk (e.g., attributes, causes, effects)] per [author/source] in [domain]
A: [Concise, clear answer]
</card_2>

<card_3>
[9 words; 6 words]
Q: [Question exploring implications or connections, or a cloze deletion for lists] per [author/source] in [domain]
A: [Concise, clear answer, or full list for cloze deletion]
</card_3>
</cards>
</chunk_1>

<chunk_2>
[Similar structure as chunk_1, adapted to the specific content]
</chunk_2>

... (continue for all five chunks)
</flashcard_set>

After presenting the flashcards for the five most important chunks, list any additional chunks that could be used for flashcards if the user is interested. Format this list as follows:

<additional_chunks>
<chunk_6>
[Brief description of additional chunk 6]
</chunk_6>
<chunk_7>
[Brief description of additional chunk 7]
</chunk_7>
... (continue for all additional chunks)
</additional_chunks>

Remember to:
- Group flashcards by information chunks.
- Number each chunk.
- Provide 1-3 flashcards for each of the five selected chunks.
- Ensure the median question and answer avoid compound phrases and are between 5-15 words each.
- Use the specified format for question-answer cards and list cards.
- Include extra context in parentheses when necessary.
- Place the word count above each card, not below.
- Include author/source information and domain context in questions when relevant.

Your final output should be a set of well-crafted flashcards that effectively capture the most important knowledge from the text and promote thorough understanding. Aim for precision, conciseness, and effectiveness in promoting learning and retention.

You will analyze the following text and create flashcards based on it:

Format

Could you now delete all the formatting text and just have the content?