The Asterism approach

Writing is a craft, and to each craft its tools.

In this essay we’ll lay down an iterative framework that goes beyond prompt engineering as we seek to reconcile human intuition and machine reasoning in the process of communicating ideas in writing. Leaning into what humans and LLMs do best — our capacity for original thought, and their ability to process vast datasets to reveal insightful patterns — we’ll introduce the Asterism approach: LLM-assisted information cluster articulation in argumentative writing.

“…in argumentative writing, an asterism is a self-contained cluster of information that contributes to the overall story.”

In astronomy, these areas are called asterisms: recognizable patterns of stars forming parts of a larger constellation. Take the Big Dipper in Ursa Major for example. The analogy here is that, in argumentative writing, an asterism is a self-contained cluster of information that contributes to the overall story. Good essays have well-defined information clusters: self-contained conceptual frameworks built around words speaking to the author’s perspective and intent. The Asterism Approach guides writers in leveraging LLMs to clearly articulate core concepts during the exploratory phase of their writing process, leading to focused information clusters that enrich their overall narrative.

“During the exploratory phase of your writing process, prioritize prompts that maintain the focus on the micro-level, rather than soliciting broad overviews.”

Shaping conceptual frameworks first gives us the clarity to iterate and refine our understanding of the subject matter we are exploring, allowing us to steer our story towards results that align with our original intention and intuition. LLMs reasoning capabilities are determinant in this exploration phase where we investigate what we are talking about, how we are talking about it, and why we are talking about it. In what follows, we’ll go over the tasks, workflows, and prompts needed to articulate concepts and shape focused information clusters.

Structured loops of refinement where raw thoughts evolve into precise articulation through systematic revisions. Contains tasks, and workflows.

a. Vocabulary: Expand and refine vocabulary for precise articulation.

b. Figurative Language: Study the use of literary devices (metaphors, similes) to help you illustrate the concepts you are discussing.

c. Identify Core Ideas: Extract keywords and main ideas from your thinking.

d. Visual Mapping: Visually organize keywords/sentences to structure ideas on a piece of paper or digital canvas.

1. First Brain Upload in Notion: String together your keywords manually (no LLM) to create coherent sentences in your default text editing tool. I like Notion because I can store each iteration as a ‘toggle list’ on the same page.

2. Bring First Brain Upload to LLM: Paste your raw thoughts into your LLM of choice. I find Google AI Studio particularly useful because it doesn’t take excessive creative liberties with your text. 

• My go to prompt here is: “Articulate this [brain upload]”. The prompt’s conciseness, along with the 100% human input, ensures that the LLM dedicates its reasoning capabilities to enhancing the clarity and precision of my raw thoughts. Longer, more complex prompts dilute this focus. Introducing broader concerns, such as the essay’s overall structure, leads to a loss of precision and generic outputs.

3. Extract Successful Elements: Carefully review the LLM’s output and extract any elements that resonate with you — a well-chosen word, a compelling figure of speech, or a sentence that crystallizes an aspect of your concept particularly well. Copy these successful elements into your Notion page as individual bullet points.

4. Manually String Together Successful Elements: Using your own judgment, manually arrange the extracted elements into coherent sentences. Feel free to string sentences into paragraphs or keeping them as bullet points for now.

Steps 1 through 4 = one iteration cycle: Log each iteration cycle as its own version within a toggle list on your Notion page for easy progress tracking. Use the latest output from an iteration cycle as the starting point for the next.

5. Repeat: Work your way through several iteration cycles.

Key LLM-generated short outputs that provide clarity, guiding what comes before or after in your writing.

• The LLM generates the anchor point: “Algorithmic bias perpetuates existing societal inequalities, embedding them within the very fabric of our technological infrastructure.”

This phrase above perfectly captures the central problem. It’s concise and highlights the systemic nature of the issue.

Carefully crafted instructions that shape LLM outputs, ensuring they enhance rather than overshadow human input.

The input provided to the LLM should always be predominantly human-generated content. Consider these starting points, replacing “[xyz]” with your own writing:

Use these prompts in your different iteration cycles of the exploration phase.

In any given situation — within the intimate or the public, the friendly or the professional — exchanged information carries meaning shaped by the communicator’s intent and the recipient’s interpretation. Fundamentally, good stories are love notes we give others. In a time when we drown in information, that which delivers its meaning with precision is remarkable and memorable. Something we are thankful for: the right words to a friend in need, an email that’s worth reading, or a thought-provoking essay, speak to our ability to reconcile intuition and reason to create meaning.

focusing on early-stage concept articulation, the approach enables writers to shape focused “asterisms” — self-contained clusters of information — that form the bedrock of a compelling narrative. This method ensures that, even with the assistance of AI, the writer retains complete control over their ideas, shaping them to resonate with their original intent and intuition