Tag: prompt engineering

Categories
AI Blogs/Weblogs

How to Refresh Dated Blog Articles

Woman typing on laptop at wooden table in cafe with coffee and pastry
A woman working on her laptop at a sunny cafe table with coffee and pastry

This afternoon I used the following prompt to help me assess posts I’d written a few years ago for updating:

I have a personal blog at https://sjl.us
Please review my blog posts that are between two and five years old and give me a prioritized list of which ones you think I should update.

I was curious how well it would do at this task. Turns out I was pleasantly surprised. Here’s the result:

Here’s my prioritized list of your blog posts (published roughly April 2021–April 2024, i.e., 2–5 years old as of April 2026) that I recommend updating, ranked from highest to lowest priority.

I reviewed the site structure, monthly archives, and indexed posts via searches. Your blog has dozens of posts in this window (mostly personal essays, photography, history, productivity/PKM, and heavy AI experimentation from the 2022–early 2023 wave). I prioritized based on:

  • How much the topic has aged (AI/prompting/tech tools change fastest; photography, personal reflections, and history are more timeless).
  • Potential value of an update (adding 2026-era insights, new examples, current tool versions, or follow-up reflections makes the post fresher and more useful to readers).
  • Thematic clustering (your AI/PKM series from 2022–2023 forms a natural arc that would benefit from a cohesive refresh).

I focused on the highest-impact ones rather than every post. Many others (e.g., pure street photography, freeway essays, family stories like the Tracy obituary, or 2021 COVID-era reflections) are still strong as-is and don’t need updates unless you want to add new photos or minor notes.

Top Priority (Strongly recommend updating soon — these are the most dated but highest-potential)

  1. Multi-step prompting for Claude (August 11, 2023)
    https://sjl.us/2023/08/11/multi-step-prompting-for-claude/
    Why update? Early Claude-specific prompting techniques; the entire field (Claude 3/4 families, GPT-4o-class models, Grok, etc.) has advanced dramatically. Easy win: expand with 2026 best practices, new multi-step examples, and comparisons.
  2. My Current PKM Stack as of September 2023 (September 20, 2023)
    https://sjl.us/2023/09/20/my-current-pkm-stack/
    Why update? Details your Obsidian/Instapaper/Feedly/Drafts workflow at the time. Tools and plugins have evolved a lot; readers love current “state of my stack” posts. A 2026 version would be highly engaging.
  3. Using AI to Help Write (November 15, 2022)
    https://sjl.us/2022/11/15/using-ai-to-help-write/
    Why update? Pure GPT-3-era experimentation. Perfect opportunity to contrast 2022 capabilities with today’s multimodal models and show how your own AI-assisted writing process has matured.
  4. LLMs = Dream Machines (December 11, 2023)
    https://sjl.us/2023/12/11/llms-dream-machines/
    Why update? References to early hallucination discussions and Andrej Karpathy’s tweet feel historic now. Update with current “dream machine” examples and how you use LLMs in 2026.
  5. Crafting Effective Prompts for Your AI Assistant (Claude, ChatGPT, Bard, etc.) (September 9, 2023)
    https://sjl.us/2023/09/09/crafting-effective-prompts-for-your-ai-assistant-claude-chatgpt-bard-etc/
    Why update? Another core prompting/PKM-adjacent piece from the explosion period. Pair it with the multi-step post for a nice “2023 vs 2026 Prompting Guide” refresh.

High Priority (Worth updating — clear value-add)

  1. An AI Thanksgiving (November 22, 2022)
    https://sjl.us/2022/11/22/an-ai-thanksgiving/
    Why update? Fun GPT-3 + DALL·E experiment. Readers would love a 2026 “then and now” version with newer models/image generators.
  2. Hacking Creativity with AI (August 22, 2023)
    https://sjl.us/2023/08/22/hacking-creativity-with-ai/
    Why update? Early chatbot creativity techniques; the creative workflow space has exploded since then.
  3. On the Minimum Viable Environment (MVE) and Apple’s Vision Pro (July 1, 2023)
    https://sjl.us/2023/07/01/on-the-minimum-viable-environment-mve-and-apples-vision-pro/
    Why update? Pre-launch speculation. Vision Pro shipped in 2024 — you can now add real-world usage, ecosystem updates, and whether MVE thinking held up.
  4. Drafts – a tool for idea capture (April 16, 2022)
    https://sjl.us/2022/04/16/drafts-a-tool-for-idea-capture/
    Why update? The app was celebrating its 10th anniversary then; it’s now 14+ years old with major feature/UI changes. Your long-term usage makes this a great “then vs now” post.

Medium Priority (Nice-to-have updates if you have time)

  1. Reading Books in 2022 (May 11, 2022)
    https://sjl.us/2022/05/11/how-im-reading-books/
    Why consider? Kindle/app habits and highlighting workflows have evolved; easy to refresh with current tools and your 2026 reading volume.

Other notable 2023 AI-adjacent posts (e.g., NotebookLM deep-dive from Dec 2023, Explorations of PKM from Oct 2023) would slot in here if you want to do a broader “AI/PKM 2023 Retrospective” series.

Lower Priority / Probably Leave As-Is

  • Photography series (e.g., “Capturing the Art in Everyday Scenes,” Paris posts, “Redwood Circle,” etc.) → Timeless and visual; they hold up well.
  • History/personal essays (e.g., “Ode to a Queen,” “The World’s Most Beautiful Freeway,” “Remembering Blue Chip Stamps,” “Old Man Walking,” 2021 book recs, “Stuck in Neutral”) → Evergreen or emotionally resonant; updating could dilute their original voice unless you have new material.
  • Holiday/personal posts (e.g., Thanksgiving 2023, Merry Christmas 2021) → Fine as snapshots in time.

Quick recommendation on approach: Start with the top 5 (they form a nice cluster around AI prompting/PKM). Updating them together would create a strong “2023 AI Reflections – Revisited in 2026” section or category on the blog. Many of these posts already reference each other, so cross-linking updated versions would be powerful.

If you’d like me to pull full content from any specific post, draft sample update language, suggest new sections, or help with images/AI-assisted rewrites, just let me know! Your blog has a great reflective voice — these updates would keep it feeling current without losing that personal touch.

Categories
AI

Beyond the Summary: Using AI to Find the “Friction” in Your Thinking

We’ve reached the “Summary Plateau.”

You see it everywhere. Every browser extension, every note-taking app, and every enterprise LLM now offers a “Summarize” button. It’s the ultimate promise of the efficiency era: Give us the 2,000-word essay, and we’ll give you the three bullet points. But there’s a hidden tax on this kind of efficiency. When we ask an AI to summarize, we are asking it to smooth out the edges. We are asking it to remove the “noise.” The problem is, in the world of ideas, the noise is often where the signal lives. The friction—the parts of an argument that make us uncomfortable or that we don’t quite understand—is where the actual learning happens.

If we only consume the summaries, we aren’t thinking; we’re just acknowledging.

The Mirror, Not the Maker

I’ve been experimenting with a different approach. Instead of asking the model to make the content shorter, I’ve been asking it to make my engagement with the content harder.

I don’t want a “Maker” to write my thoughts for me. I want a “Mirror” to show me where my thoughts are thin.

When I’m wrestling with a complex piece—perhaps a deep dive on the future of venture capital or a philosophical treatise on Arete—I’ve stopped clicking “summarize.” Instead, I feed the text into the LLM and use these “Friction Prompts” to find the sand in the gears:

The Essential Toolkit

  • The “Steel Man” Challenge: “I am inclined to agree with this author’s conclusion. Find the three strongest counter-arguments that this text ignores, and explain why a reasonable person would hold them.”
  • The “Recursive Logic” Audit: “Identify the three most critical ‘logical leaps’ the author makes—points where a conclusion is reached without sufficient evidence. If those leaps are wrong, how does the entire argument collapse?”
  • The “Blind Spot” Audit: “What are the underlying cultural or economic assumptions this author is making that they haven’t explicitly stated?”
  • The “Cross-Pollination” Filter: “Connect the central thesis of this article to a seemingly unrelated field (e.g., Stoic philosophy or biological ecosystems). How does the logic of this text hold up—or fail—when applied to that different domain?”
  • The “Analog Translation” Test: “If I had to explain the core mechanism of this abstract concept using only physical, analog metaphors (like plumbing or woodworking), how would I do it? Where does the metaphor break down?”
  • The “Socratic Sharpening”: “Don’t summarize this. Instead, ask me three probing questions that force me to apply the core logic of this essay to a completely different industry.”

Sharpening the Blade

Summary is about completion (getting it done). Friction is about cognition (getting it right).

When the AI points out a blind spot in an article I loved, it creates a moment of cognitive dissonance. That “click” of discomfort is the sound of a mental model being updated. It’s the digital equivalent of using a whetstone on a blade—you need the friction to get the edge.

As we move further into this age of “Flash-Frozen Cognition,” the temptation to automate our understanding will only grow. But discernment—that uniquely human trait we’ve discussed here before—cannot be outsourced to a bulleted list.

The next time you’re faced with a daunting PDF or a dense long-read, resist the “Summarize” button. Ask the machine to challenge you instead. You might find that the most valuable thing the AI can give you isn’t an answer, but a better version of your own question.

A Deep Dive (Further Reading from the Archive)

If you resonated with this piece on cultivating discernment, you might find these earlier synthesis experiments worth a revisit:

  • On Flash-Frozen Cognition: A foundational post discussing how LLMs are freezing the current consensus, and how we must resist it.
  • The Harvest and the Algorithm: Comparing 1920s ice harvesting to 2020s cognition—the critical shift from scarcity to abundance.
  • The Arete of Attention: A look at the Stoic concept of virtue as the intentional direction of our most scarce resource: focus.
  • Longhand Thinking: Why the physical act of writing is the ultimate antidote to digital velocity.
Categories
AI YouTube

From Paperbacks to YouTube

Yesterday while reading the Sunday New York Times I came across the Book Review section and its cover story about paperback books. I have such fond memories of shopping for paperbacks at local drug stores, bookstores, etc but those days are long gone.

The cover story included a photograph of a stack of paperback books. I’ve learned in the past how useful AI can be at quickly turning a photograph of a shelf or stack of books into a list of titles and authors – so natually I snapped a photo of the stack of books and asked Gemini:

What are the titles and authors of these paperbacks?

Gemini came back with the list of titles and authors:

This image features a collection of vintage-style paperback spines, often found in artistic book-themed prints. The spines represent a mix of classic literature, science fiction, and pulp fiction.

Below is the list of titles and authors as they appear from top to bottom…

There were a few titles in the photo that were impossible for me to make out. I asked Gemini about them:

Were there some you couldn’t read?

Gemini came back:

I did have to do a little detective work on a few of those! While most were clear, some of the older pulp titles have very tiny or weathered text that required a bit of cross-referencing to confirm.

….

A Fun “Easter Egg”

You might notice that J.D. Salinger and Henry Kane appear twice. This stack isn’t a random collection; it’s actually a popular art piece (often titled The Librarian’s Daughter or similar) that mixes “high” literature like Middlemarch with “low” pulp fiction like Affair for a Corpse to show the breadth of a true reader’s shelf.

Continue reading “From Paperbacks to YouTube”
Categories
AI AI: Large Language Models

The Echo Effect: Why Prompt Repetition is AI’s Best Kept Secret

In our relentless pursuit of complexity, we often overlook the elegant simplicity of a fundamental human habit: repeating ourselves.

We build colossal architectures, weave intricate neural networks, and throw mountains of computational power at our artificial intelligence systems, hoping to squeeze out a few more drops of reasoning and logic. Yet, sometimes the most profound breakthroughs require no new code, no additional latency, and no extra training data.

Sometimes, you just have to say it twice.

In a fascinating December 2025 paper titled “Prompt Repetition Improves Non-Reasoning LLMs,” researchers Yaniv Leviathan, Matan Kalman, and Yossi Matias uncovered an almost absurdly simple “free lunch” in AI optimization.

Their premise is straightforward: when you aren’t using a heavy reasoning model, simply copying and pasting your input prompt multiple times significantly boosts the model’s performance.

“When not using reasoning, repeating the input prompt improves performance for popular models (Gemini, GPT, Claude, and Deepseek) without increasing the number of generated tokens or latency.”

The mechanics behind this are elegantly pragmatic.

By repeating the prompt, you are moving the heavy computational lifting to the parallelizable “pre-fill” stage of the model’s processing. The AI’s causal attention mechanism gets to process the same tokens again, allowing the later iterations of the prompt to attend to the earlier ones. It effectively acts as a hack to simulate bidirectional attention in a decoder-only architecture.

What’s even more telling is the paper’s observation on why this works so well.

The researchers noted that models trained with Reinforcement Learning (like OpenAI’s deep-thinking variants) naturally learn to “restate the problem” in their internal monologue. They figured out on their own what these researchers are suggesting we do manually: repeat the question to focus the mind.

Reading this paper, I couldn’t help but draw a parallel to the human condition and the nature of listening.

How often do we assume that because we have articulated a thought once, it has been fully absorbed? We fire off a single, dense instruction to a colleague, a partner, or a friend, and then marvel when the nuance is lost in translation.

We suffer from our own attention bottlenecks.

Like a non-reasoning LLM trying to parse a complex query in a single pass, we are constantly bombarded with a stream of tokens—emails, notifications, conversations, fleeting thoughts. To truly understand, to truly digest and synthesize information, we need the grace of repetition.

There is a strange poetry in the fact that to make our most advanced digital minds smarter, we have to talk to them the way we talk to a distracted child or a busy spouse. The “microscope effect” highlighted in the study—where repeating a prompt drastically improved extraction tasks—shows that the failure wasn’t in the model’s capacity to know, but in its capacity to focus. Repetition forces focus. It creates a resonant echo in the context window, a digital highlighter that screams, “This matters. Look here again.”

As we continue to navigate a world increasingly augmented by artificial intelligence, this paper serves as a humbling reminder. The bleeding edge of technology isn’t always found in the most complex equation; sometimes, it’s hidden in the most basic principles of communication.

Whether you’re prompting a billion-parameter language model or trying to connect with the human sitting across from you, the lesson is clear.

Clarity isn’t just about the words you choose. It’s about giving those words the space, the resonance, and the repetition they need to be truly understood.

Say it once to be heard; say it twice to be understood.

Categories
AI Work

The Centaur’s Dilemma: What Chess Teaches Us About the AI Era

Note: this post was stimulated by a recent conversation between Dario Amedei and Ross Douthat.

In 1998, Garry Kasparov did something unexpected after his historic defeat to IBM’s Deep Blue: he teamed up with the machine. He pioneered “Centaur Chess,” a hybrid format where human intuition merges with cold, silicon calculation. The human acts as the executive, the engine as the raw horsepower. For a time, it was the highest level of chess ever played.

But there is a sobering lesson hidden in the evolution of this game. We are currently living through the workforce equivalent of the Centaur era, and history suggests our “hybrid honeymoon” won’t last forever.

Right now, we are in the augmentation phase. A junior copywriter or coder armed with a Large Language Model can suddenly produce work at a staggering pace. The AI acts as a great equalizer, much like a mediocre chess player with a strong engine beating a Grandmaster in the early 2000s. We are shifting into executive roles—prompting, curating, and orchestrating rather than creating from scratch.

However, in modern Centaur Chess, a chilling reality has emerged: human intervention now yields negative returns. The engines have become so impossibly advanced that when a human overrides Stockfish today, they are almost certainly making a mistake. The human loop, once the ultimate strategic advantage, has become a liability.

This is the “Grandmaster Floor” problem, and it is coming for the job market.

“Eventually, companies may view human oversight not as a ‘value add,’ but as an insurance cost they’d rather cut.”

We are seeing this fracture already. Pure “engine” industries—entry-level data analysis, logistical tracking, basic customer support—are rapidly phasing out the human element because human latency is a drag on the system. Yet, in fields requiring high-stakes moral judgment or empathy, like healthcare or law, the Centaur model remains deeply necessary.

This forces a deeply personal question: How do we stay relevant when the engine eventually solves the game?

The answer lies in recognizing the boundaries of the board. Chess is a closed, finite system. Human life and business are open, messy, and infinitely complex. The survival strategy isn’t to compete on calculation, but to double down on connection, empathy, and problem definition. AI is brilliant at providing the perfect answer, but it fundamentally lacks the soul to know which questions are worth asking.

In the future, the human touch won’t just be a necessity; it will be a luxury. The most valuable skill won’t be navigating the engine, but deciding where the engine should go.

A couple of considerations:

• Take an honest look at your daily work: how much of your time is spent “calculating” (tasks an engine will soon do better) versus “evaluating” (deciding what actually matters)?

• If the technical, process-driven aspects of your job were completely automated tomorrow, what uniquely human value—empathy, context, or connection—would you still bring to the table?

Categories
AI AI: Prompting Prompt Engineering

Using AI to Improve Prompts

This morning I came across a prompt that I’ve been finding provides me with some interesting results in understanding the financial results of a public company. I’ve used it several times on various companies and have found new insights in the results that I otherwise might not have realized. Here’s the prompt:

Analyze the last 3 earnings calls for [symbol].

Don't summarize what they said - tell me what they're NOT saying.

Focus on:

1) Questions the CEO dodged or gave vague answers to,
2) Metrics they stopped reporting compared to previous quarters,
3) Language changes - where they went from confident to cautious or vice versa,
4) New talking points that appeared suddenly,
5) Guidance changes and the exact wording they used to frame it. Then connect this to their stock performance in the 2 weeks following each call.

What pattern emerges? Explain clearly. 

Is the stock a sell or a buy at its current price? How strong is your conviction?

One of the techniques that advanced AI users employ is to use AI to improve on their prompts. I wanted to see what Gemini 3 Pro could do to help me improve on this particular prompt – so I asked it:

You are an expert at prompt engineering. Here's a prompt that I find very useful but would like help to improve. What would you recommend to make it even better? Prompt: [insert prompt here]

Gemini responded and educated me:

This is a sophisticated request. You are asking for negative space analysis—detecting the absence of signal, which is significantly harder for LLMs than summarizing present text.

The weakness of your current prompt is that it relies on the model’s general “feeling” about what was dodged. To make it “expert-level,” you need to force the model to perform comparative forensics between the Prepared Remarks (scripted/safe) and the Q&A (unscripted/risky), and between historical promises and current reality.

Here is the recommended upgrade to your prompt, followed by an explanation of why these changes work.

And the suggested prompt it provided when I asked it to optimize the prompt for $PYPL:

Role: Act as a Forensic Financial Analyst and Researcher. Task: Analyze the earnings call transcripts for $PYPL (PayPal) from Q1 2025, Q2 2025, and Q3 2025.

Core Directive: Do not summarize the "Bull Case." Your goal is to identify omission, deflection, and narrative drift.

Step 1: The "Dog That Didn't Bark" (Metric Analysis)

Compare the metrics reported in the Q1 2025 deck/opening remarks against Q3 2025.

List any specific metric (e.g., Net New Actives, TPV ex-eBay, OCF growth) that was prominent in earlier calls but vanished or was de-emphasized in the most recent call.

Output: A table: [Metric Name] | [Q1 Mention Count] | [Q3 Mention Count] | [Context of Disappearance].

Step 2: Semantic Divergence (Prepared vs. Q&A)

Analyze the tone difference between the CEO's Prepared Remarks (Scripted) and their Q&A Responses (Unscripted).

Identify specific instances where a confident claim in the script was walked back or qualified with "hedging" words ("essentially," "broadly," "we believe," "monitor") during the Q&A.

Step 3: The "Non-Answer" Detection

Review the Q&A section. Identify questions where the executive:

Pivoted to a different topic (e.g., asked about margins, answered about volume).

Used "word salad" (high jargon, low specific detail).

Claimed it was "too early to tell" for a product launched >6 months ago.

Quote the specific question and the specific non-answer.

Step 4: Stock Correlation

Retrieve the stock price performance for the 10 trading days following each call.

Correlate the severity of the "evasiveness" you found in Step 3 with the subsequent price action.

Final Verdict:

Based only on the negative signals (omissions and deflections), rate the stock as Accumulate, Neutral, or Distribute (Sell).

Provide a "Conviction Score" (1-10) based on the clarity of the executive team's transparency.

Gemini went on to explain the enhancements it made including the suggesting prompt for the Gem:

Why This is Better

  1. It defines the “Negative Space” (Step 1) LLMs hallucinate less when you ask them to count. By asking for a specific comparison of mention counts of metrics between Q1 and Q3, you force the model to mathematically prove that a topic has been dropped.
  2. It separates Script from Reality (Step 2) Executives are media-trained to sound confident in their opening speech. The “alpha” (the edge) is found when they get flustered in the Q&A. This prompt explicitly instructs the model to treat those two sections differently.
  3. It targets specific evasion tactics (Step 3) Instead of just asking for “dodged questions,” this prompt gives the model a taxonomy of evasion: Pivoting and Word Salad. This helps the model recognize that “We are excited about the long-term trajectory of Venmo” is actually a non-answer to “Why did Venmo revenue drop this quarter?”
  4. It forces “Conviction Scoring” Asking for a simple “Buy/Sell” often yields a hedged, safe answer from AI. Asking for a “Conviction Score” forces the model to weigh the evidence. If it finds many dodged questions, the conviction score for a “Sell” should naturally be higher.

This is just one example of using AI to enhance what’s possible using AI. I found this to be a very interesting and educational exercise and example!

Various AI’s natually respond differently to this prompt – and it’s interesting to compare the results for the same company. Give it a try!

After running the prompt one a couple of companies, I realized that this is a prompt I’d like to generalize and save away as a Gemini Gem so that I can easily run it whenever I want. So I asked Gemini:

I would like to generalize this and turn it into a gem that I can run on any company.

Gemini’s response:

This is a great move. By generalizing this, you are creating a “Forensic Short-Seller” agent that you can drop onto any earnings situation.

Below is the Master Prompt Template. I have swapped out the PayPal-specific metrics for variable placeholders and generalized the logic to work across industries (e.g., it will look for “Same Store Sales” in retail just as aggressively as “ARR” in software).

The “Negative Space” Forensic Prompt
Copy and paste the text below. All you need to do is replace the bracketed [VARIABLES].

Role: Act as a Forensic Financial Analyst and Short-Seller Researcher.

Context: I need you to perform a “Negative Space Analysis” on the last 3 earnings calls for [INSERT TICKER SYMBOL].

The Theory: Corporate executives are incentivized to highlight growth and obscure deterioration. The truth is often found in what they stop saying, or how they change their phrasing, rather than their headline numbers.

Your Goal: Do not summarize the “Bull Case.” Identify omission, deflection, and narrative drift.

Step 1: The “Vanishing Metric” Analysis

Scan the “Key Performance Indicators” (KPIs) reported in the oldest of the three calls.

Compare them to the most recent call.

Identify any specific metric (e.g., Customer Count, Retention Rate, Margins per Unit, backlog) that was prominent in earlier calls but was removed, redefined, or de-emphasized in the most recent call.

Output: A table: | Metric Name | Mention Count (Oldest Call) | Mention Count (Newest Call) | The Executive Excuse (if any given) |

Step 2: Semantic Divergence (Scripted vs. Unscripted)

Analyze the tone difference between the CEO/CFO Prepared Remarks (Scripted) and their Q&A Responses (Unscripted).

Identify instances where a confident claim in the script was walked back in the Q&A using “hedging” language (e.g., “headwinds,” “macro environment,” “transitory,” “conservative view”).

Flag any “Adjective Inflation”—using strong adjectives (“robust,” “tremendous”) to describe mediocre data.

Step 3: The “Non-Answer” Detection

Review the Q&A section specifically. Identify questions where the executive:

Pivoted: Asked about A, answered about B.

Filibustered: Gave a long, jargon-heavy answer without specific numbers.

Deferred: Claimed it is “too early to tell” for a strategic initiative that should have data by now.

Format:

The Question: [Quote the Analyst’s hard question] The Evasion: [Quote the specific part of the answer that dodges] Translation: [Briefly explain what they were hiding]

Step 4: Sentiment vs. Price Action

Look at the stock price performance in the 10 days following each of the 3 calls.

Compare the CEO’s confidence level (High/Medium/Low) with the market’s reaction.

Pattern Recognition: Does the market consistently sell off despite “good news” (indicating management has lost credibility)?

Final Verdict:

Rating: Accumulate / Neutral / Distribute (Sell)

Conviction Score (1-10): (10 = The company is totally opaque/hiding bad news; 1 = The company is transparent).

The “Smoking Gun”: One sentence summarizing the single most concerning omission or change in narrative.

How to use this “Gem” effectively
To get the most out of this generalized prompt, keep these three tweaks in mind:

Define the “Golden KPI”: If you know the industry, add a hint in Step 1.

For SaaS: “Focus on Net Dollar Retention and ARR.”

For Retail: “Focus on Same Store Sales and Inventory levels.”

For Manufacturing: “Focus on Order Backlog and Book-to-Bill ratio.”

Provide the Data: Most LLMs cannot browse live transcripts behind paywalls perfectly. For best results, paste the text of the transcripts (or the Q&A sections) directly into the chat before running the prompt, or ensure you are using a model with active web browsing that can access financial news sites.

The “Competitor Check”: If you want to make it even sharper, add this line to Step 2: “Contrast their tone with [Competitor Ticker], who reported recently.”

Categories
AI AI: Prompting

Carving Away: Part II

Well, that escalated quickly.

It looks like the wood carving prompt from my recent post struck a chord. I’ve seen some incredible results floating around, and since traffic is still high, I wanted to share a few “remixes” of the original prompt that I’ve been experimenting with this weekend using Gemini 3 Pro.

If you mastered the basic “hand-carved miniature” look, here are three ways to push the aesthetic in different directions.

1. The “Dark Walnut” Aesthetic

The original prompt tends to produce a light, pine-like wood. Use this variation if you want something moodier, richer, and more polished. It works exceptionally well for portraits or architectural subjects where you want a high-end feel.

The Prompt Addition:
Append this to your subject line: ...carved from dark polished walnut wood, rich deep grain texture, rim lighting, subsurface scattering, mahogany tones, smooth finish.

2. The “Painted Folk Art” Look

Sometimes raw wood is a bit too monochromatic. This variation pushes the model to apply a distressed paint job, making the image look like a vintage toy or traditional folk art found in an attic.

The Prompt Addition:
Append this to your subject line: ...faded hand-painted wooden figurine, chipped paint revealing wood underneath, vintage folk art style, muted primary colors, distressed texture.

3. The “Rough Hewn” Sketch

This is my personal favorite. It forces the AI to leave “tool marks,” making the object look like a work-in-progress rather than a finished product. It adds a tactile, human imperfection that many AI images lack.

The Prompt Addition:
Append this to your subject line: ...rough hewn unfinished wood, visible chisel marks, splintered edges, raw timber texture, sawdust specs, harsh directional lighting.

A Note on Models

I am still finding that Gemini 3 Pro handles the texture mapping better for the “Rough Hewn” look, capturing the jagged edges convincingly. However, Nano Banana seems to prefer the polished “Dark Walnut” style, producing really nice specular highlights on the wood grain.

If you are getting results that look too “plastic,” try lowering the guidance scale slightly or adding varnish to your negative prompt.

Let me know what you carve out of the latent space next.

Categories
AI AI: Prompting Writing

AI as a Mirror, Not a Maker

I’ve been thinking a lot lately about how we move past the novelty phase of AI—beyond just asking a chatbot to “write a poem about a turkey” or summarize a meeting—and into actual thinking with these tools.

As a lifelong learner, I’m always on the hunt for workflows that help me synthesize information better. Most of the “AI for writing” advice I see online is pretty generic. But I recently came across a breakdown of how four high-profile writers are making effective use of tools like NotebookLM and Claude in ways that are much more sophisticated than simple text generation.

What jumped out at me is that none of these writers use AI to write for them. They use it to structure, challenge, and code.

Here are the four models that caught my eye.

1. The Triangulated Research Base (Steven Johnson)

Steven Johnson (Where Good Ideas Come From) has a workflow that solves a problem I face constantly: the messy “research phase.”

Instead of treating the AI as an oracle, he treats it as a connection engine. He creates a dedicated notebook (using Google’s NotebookLM) and uploads three distinct types of sources: a primary source (like a raw PDF or study), a secondary source (like a context article), and a multimedia transcript.

Then, rather than asking for a summary, he asks the AI to find the friction between them: “What themes appear in the interview transcript that contradict the historical account in the PDF?”

It’s less about getting an answer and more about finding the blind spots in your own reading.

2. The Diagnostic Editor (Kenny Kane)

This one really resonated with me because it mirrors the experiment I tried recently with my “Bubble Bath” post.

Kenny Kane uses Claude not to generate prose, but to act as a ruthless developmental editor. He uploads a messy draft and runs a “Diagnostic” prompt. He doesn’t ask “fix this,” he asks: “Where does the argument drift? Where does the energy drop?”

He even has the AI analyze his best writing to identify his specific “DNA” (sentence length, vocabulary choice) and then asks it to apply that same tone to his rougher sections. It’s using the AI as a mirror rather than a ghostwriter.

3. The Memo-to-Demo Shift (Dan Shipper)

Dan Shipper at Every is doing something fascinating that changes the definition of writing altogether. He argues that in the AI age, we shouldn’t just describe a concept; we should build a small app to demonstrate it.

If he’s writing about “Spaced Repetition,” he doesn’t just explain the theory. He asks Claude’s Artifacts feature to “Write a React component that lets a user test spaced repetition live in the browser,” and then embeds that little app directly into the essay. The writing becomes 50% prose and 50% software.

4. The Co-Intelligence Loop (Ethan Mollick)

Ethan Mollick focuses on breaking the echo chamber. Before he publishes, he spins up simulated personas—a skeptical VC, a confused novice, an expert in a tangential field—and asks them to critique his draft from their specific viewpoints.

It’s effectively a focus group of one.

How to Get Started

If you’re like me, seeing all these workflows might feel a bit overwhelming. My advice? Don’t try to overhaul your entire writing process overnight. Just pick one experiment to try this week.

Here are two simple entry points:

Experiment A: The “Blind Spot” Check (For Research)

If you are reading up on a topic, don’t just take notes. Open Google NotebookLM, create a new notebook, and upload your sources (PDFs, URLs, or pasted text). Then, ask this specific question:

“Based strictly on these sources, what is the strongest argument against my current thinking? What connection between Source A and Source B am I missing?”

Experiment B: The “Ruthless Editor” (For Writing)

If you have a rough draft sitting on your hard drive, copy it into Claude or ChatGPT and use this prompt (adapted from Kenny Kane’s workflow) before you do any manual editing:

“Act as a senior editor. Do not rewrite this text. Instead, analyze my draft and tell me: 1) Where does the argument lose energy? 2) Does the opening hook successfully promise what the conclusion delivers? Be critical.”

I’ve found that using the tools this way—as a partner for thinking rather than just generating—is where the real magic happens.

Which one will you try first?