Hey everyone!

This week, I’m sharing a failure. My plan to automate content with Notion completely fell apart, and I want to show you exactly what I learned from it.

Thank you for all the great feedback so far—it’s been incredible. Sharing the wins is easy, but I think it’s just as important to share the stumbles.

Also in this issue:

• AI Education: The difference between structured and unstructured data (and why it’s the key to better AI results).
• A 10-Minute Win to compare and contrast two ETFs in minutes.

Let's dive in with your curated AI news from the past week. 

Missed a previous newsletter (released every Tuesday at 9AM EST)? No worries, you can find them on the Archive page at MindOverMoney.ai. Don’t forget to check out the Prompt Library, where I give you templates to use in your AI journey. If you have content suggestions, a workflow idea, or want to share a success from your AI journey, reach out at admin@mindovermoney.ai.

1) Google’s Gemma helps flag a new cancer-therapy pathway

Summary: Google details how a Gemma-based model (C2S-Scale) predicted a combo (silmitasertib + interferon) that boosts immune signaling in “cold” tumors; wet-lab tests validated the effect, pointing to a promising treatment direction.

Why it matters: Clear, beginner-friendly example of AI accelerating biomedical discovery—turning model hypotheses into lab-verified leads that could inform future oncology research.

2) Top U.S. Army general says he’s letting ChatGPT make decisions

Summary: The U.S. commander in South Korea quipped he’s grown “really close” with ChatGPT—comments that sparked heated debate about AI advice vs. military decision-making.

Why it matters: A vivid snapshot of AI’s cultural moment—and a reminder to set clear guardrails for where AI fits (and doesn’t) in high-stakes decisions.

3) Microsoft rolls out major Windows 11 AI upgrades for Copilot

Summary: Windows 11 adds “Hey Copilot” voice activation, expands Copilot Vision, and trials Copilot Actions that can book or order directly from the desktop under user-granted permissions.

Why it matters: AI is moving from a chat box into the OS itself—making everyday PCs feel more like assistants for simple automations and on-screen help.

4) OpenAI pauses MLK cameos after backlash over Sora deepfakes

Summary: Following complaints from the King estate, OpenAI suspended MLK appearances in the Sora app and says estates of historical figures can opt out—tightening guardrails on AI-generated likenesses.

Why it matters: A fast-moving, real-world example of AI policy and safety in action—useful context for creators, brands, and educators navigating rights and reputation.

5) Google AI Studio gets a cleaner playground, usage dashboards, and Maps grounding

Summary: Google rolled out an AI Studio overhaul: a unified playground for Gemini/GenMedia/TTS/Live, a real-time rate-limit/usage page, and options to ground models with Google Maps.

Why it matters: Friendlier tooling lowers setup time for first projects; seeing limits and grounding data clearly helps beginners ship useful, reliable AI apps faster.

Models don’t “think”—they learn patterns from data. Data is simply recorded facts about something that happened: numbers, words, dates and times, and context about where it came from. Let’s focus on a personal budget to make this information tangible and related to your personal finances. Budget data includes each purchase amount, the merchant, the date and time, the payment method, and any notes or line items. For language-based systems, data also includes full sentences from receipts, statements, and emails—because those sentences carry the patterns a system learns to read and write.

Good budget data tends to be: accurate (matches reality), consistent (fields mean the same thing every time), complete enough (weekdays and weekends, online and in-store), timely (new subscriptions show up), representative (covers groceries, rent, rideshare, and seasonal spikes), and described (short notes that define fields so meanings don’t drift).

Keep this context in mind as you read on; we’ll use them to examine structured, semi-structured, and unstructured data—how each becomes machine-readable, how it feeds learning, and how quality choices raise or lower accuracy in something as familiar as a household budget.

One scenario all the way through: building your monthly budget

You want a clear picture of where your money goes. Your sources match the three data “shapes”:

1. Bank export — a spreadsheet of transactions (structured).
2. Email receipts — messages with labeled headers plus free-text bodies (semi-structured).
3. PDF statements and photos of paper receipts — rich but free-form files (unstructured).

We’ll use this same budget scenario for every example below.

Structured data — “spreadsheet neat” 

Plain English: Rows and columns with fixed field names—like date, merchant, amount, category. You always know what each column means.

Budget example:

• You export transactions.csv from your bank. Each row is one purchase.
• Columns are consistent: date, merchant, amount, channel (online/in-store), category.
• An AI model can learn straight from these columns: repeated amounts on the same day each month often point to subscriptions; combinations like merchant + time of day + amount help suggest the right category.

Watch-outs (budget-specific):

• Units drifting: If “amount” switches from dollars to cents mid-file, your totals explode. Keep units steady.
• Missing slices: Only seeing weekday transactions hides weekend patterns—budgets skew low on restaurants and rides.

Semi-structured data — “has tags, not rigid”

Plain English: There are labeled parts (headers like From, Date, Total), but the body text varies by sender. Not one strict table, yet pieces are clearly marked.

Budget example:

• A grocery delivery email has merchant name and total in the header, plus a free-form list of items in the body.
• A rideshare email has a labeled pickup time and fare, followed by a flexible trip summary.

Watch-outs (budget-specific):

• Different senders use different words for the same idea (Total, Amount, Charge). Make a tiny “translation list” so they map to one Amount column.
• Treat headers as reliable anchors and the free text as supporting detail—don’t weigh them equally.

Unstructured data — “free-form”

Plain English: No fixed columns—documents, PDFs, screenshots, images, recorded calls, video. Rich in detail but not immediately searchable.

Budget example:

• A PDF bank statement for March.
• A photo of a paper receipt from a weekend farmers market.
• A scanned utility bill with line items and due date.

How it becomes useful for your budget:

• First make it readable: convert PDFs and photos into selectable text so a system can “see” the merchant, total, and date.
• Once readable, it can summarize, categorize, and route: send receipts to one folder, statements to another, highlight unusual charges or new fees.

Watch-outs (budget-specific):

• If you skip the conversion, the system learns nothing from these files.
• Blurry photos or low-quality scans cause misread numbers; better captures improve accuracy.

How an AI system “sees” your budget data

• With the spreadsheet, it compares the same fields across many rows to find stable patterns: subscriptions, typical amounts by merchant, odd spikes.
• With the emails, the labeled headers act like signposts—Merchant, Total, Date—and the body adds nuance (what you bought), which helps with category suggestions and notes.
• With PDFs and photos, a short “make it readable” step unlocks the content so it can be searched, summarized, or sorted (invoice vs. receipt vs. statement).

Why quality beats quantity

• Wrong answers teach the wrong lesson. If your “category” notes are inconsistent (sometimes “groceries,” sometimes “food”), the system learns confusion—and your budget summary drifts.
• Coverage matters. Leave out weekend cash purchases and you’ll understate restaurants and local markets. Include both channels (online/in-store) and both time slices (weekdays/weekends).
• Readable beats raw. A PDF that hasn’t been converted to text is invisible to a system. A quick conversion turns clutter into usable information.

A simple mental model to keep

1. Shape → approach:
   • Structured (spreadsheet) feeds learning directly.
   • Semi-structured (tagged emails) gives you reliable fields and helpful free text.
   • Unstructured (PDFs, photos) needs a quick “make it readable” step first.
2. Meaning → stability:
   • Write a one-line definition for each field: Amount is always dollars, Date is the transaction date, Merchant is the seller’s name.
   • Keep formats steady so totals and trends don’t drift.
3. Quality → trust:
   • Correct categories, broad coverage, and readable files build a budget you can rely on.
   • Errors, gaps, or unreadable documents quietly break it.

Enjoy a video overview created by NotebookLM.

📊 ETF Comparison in Minutes

Why this matters: Most new investors start with ETFs—but picking between “similar” funds can be confusing. In about 10 minutes, you can pull official fund facts (expense ratio, index, holdings), drop them into a free AI, and get a clean side-by-side table + plain-English “best fit for me” summary.

Step 1 — Pick 2 to 3 ETF tickers to compare (2 minutes):

Use any free source to find the funds you want:

• Yahoo Finance ETF Screener (Free): filter by category (e.g., US large cap, Tech, Dividend) to get candidate tickers.
• Issuer Lists (Free): browse Vanguard and iShares ETF lists to find funds that match your goal (total market, S&P 500, international, bonds).
  • Examples: VTI (US total market), VOO/IVV (S&P 500), VT (global stocks). Issuer pages and fact sheets are always free and most accurate.

Step 2 — Open each fund’s official fact sheet (3 minutes):

For every ticker, open the issuer’s fact sheet (PDF or webpage). You’ll find:

• Expense ratio (your ongoing fee)
• Index tracked / strategy
• Top holdings & sector weights
• AUM, inception date
• Distribution (dividend) yield
  • Example: iShares IVV (S&P 500) shows a 0.03 % expense ratio and detailed sector weights. Vanguard and iShares post updated PDF fact sheets on their websites for every ETF.

Step 3 — Use a free AI to build your comparison (3–4 minutes):

Choose any of these free AI helpers:

• Gemini (Free)
• Perplexity (Free)
• ChatGPT (Free)

Once you select your AI tool:

• Start a new chat
• Upload the PDF fact sheet of copy/paste the information from the fact sheet
• Copy/paste the prompt below -- add the tickers that the bottom

🧠 Copy + Paste Prompt 

(Works with any ETFs — just swap the tickers at the bottom.)

You are my AI investment assistant. I want to quickly compare exchange-traded funds (ETFs) side-by-side to help me make smarter, lower-cost investment decisions.

Goal:

Help me analyze and compare multiple ETFs based on official facts—like expense ratio, holdings, performance focus, and risk profile—and summarize which might fit different investor goals.

Instructions:

1. Create a comparison table for the ETFs listed below.

2. Use columns:

   - Ticker / Name

   - Expense Ratio (%)

   - Index / Strategy

   - Top 5 Holdings

   - Sector Tilts (Top 3 Sectors)

   - AUM (Assets Under Management)

   - Dividend Yield (%)

   - Inception Date

   - Notable Differences / Highlights

3. After the table, write:

   - "So What?" Section: 3–5 concise bullets explaining what each ETF is best suited for (e.g., long-term core, dividend focus, international exposure).

   - "Red Flags" Section: list any concerns (e.g., higher fees, small fund size, narrow exposure, overlap).

   - "Summary Recommendation:" explain which ETF may be the better choice for different investor profiles (e.g., beginner, income-seeker, global investor).

4. Use only the official data from each ETF’s fact sheet (not market commentary).

5. If a data point is missing, write "Not disclosed."

ETFs to compare:

VOO (Vanguard S&P 500 ETF)

IVV (iShares Core S&P 500 ETF)

SPY (SPDR S&P 500 ETF Trust)

Optional follow-ups:

“Add the official fund websites and fact sheet links as a final section.” “Explain how a 0.03 % vs 0.09 % expense ratio impacts returns over 10 years on $10 000.”

💡 You can replace the example tickers with any ETFs you want to compare—the prompt works universally.

Step 4 — Quick sanity check & tie-breaker (1 minute):

• Verify expense ratios & index names against the PDF fact sheets.
• Tie-breakers: lower expense ratio, broader index coverage, and higher AUM (liquidity means tighter spreads and less slippage).

The Payoff: 

In about 10 minutes you’ll have a side-by-side ETF table and a short plain-English summary so you can invest confidently and avoid paying more than you need.

💡 Pro tip: Save your comparison table by goal (“US Core,” “Intl Core,” “Bond Core”) and refresh it quarterly.

Transparency & Notes for Readers:

• All tools are free: Yahoo Finance Screener, Vanguard and iShares fact sheets, and Gemini / Perplexity / ChatGPT free tiers (all tested and available as of this issue).
• Data quality: Issuer fact sheet = source of truth. Use AI for formatting and explanation only.
• Not investment advice: Educational workflow to help you compare options.

👉 Your Turn:

Pick ETFs you’re considering (e.g., VOO, IVV, VTI, VT). Pull their fact sheets, run the prompt, and see what investment best suits your needs. 

My vision was clear → a fully automated tech stack that operates like a one-button newsroom. One Notion database feeding three of the four sections of Neural Gains Weekly: ideas generated, polished content created, social media posts constructed. Hit a button, and it’s done. Low cost, minimal friction, and I’m still the final editor. I thought this would be achievable and easy to execute, knowing I wasn’t going to push back the launch date of 9/30/2025. Was I naive? Did I overestimate the complexity? Did social media influence my decision-making? The answer is yes, this ‘simple’ automation build became a roadblock in creating Volume 1 by the 30th. I learned several lessons along the way that will help you build with AI.

Lesson 1 - Learning new tools takes time

I had zero experience with Notion and frankly, hadn’t researched thoroughly to understand the capabilities on this platform. I was swayed by social media hype and assumed it would be easy to build content with one click. I spent a few hours building a Notion template and trying to automatically create content. The same template behaved differently across sections. Minor changes in Notion fields broke the pipeline. Rich text turned messy. I spent hours “troubleshooting the seemingly easy parts”. The more I pushed, the more I realized the problem wasn’t the platform or formatting— it was knowledge. I needed to spend time learning and experimenting with Notion before deploying, especially given the time constraints. Too much complexity without the right knowledge base delayed my progress, which ultimately led to the realization that learning AI tools takes time and failure is part of the process.

Takeaway: Don’t automate what you haven’t learned yet. Learn the tool first, then automate the parts you truly understand.

Lesson 2: Automation is a multiplier — after the baseline is stable

I treated automation as a shortcut and a time-saver. Balancing family, a W2 career, real estate ventures, health goals, and social activities requires a strong focus on time management. The appeal of automation, especially with available AI tools, is a ‘no-brainer’ for a project like this, one requiring time, energy, and commitment to deliver on a weekly basis. I quickly learned that automation is the end state, not the beginning. I needed to build the skills and knowledge to successfully execute a complex (at least for me) automation process aligning to my vision. My focus shifted to building out a content roadmap and creating useful and engaging content. This will ensure my automation roadmap helps me accelerate progress without disrupting the momentum I have built so far.

Takeaway: Automation speeds up what already works. Build the rhythm first; automate to amplify it—not replace it.

Lesson 3: Failure leads to new ideas

The great thing about AI is the speed at which new products are released. I started the initial automation part of the project back in early August and there have been dozens of changes to the AI landscape that will benefit me during round 2 of the automation build out. My initial failures in this space were disappointing, but there is always a silver lining. I have new ideas to explore and developed a better understanding of how to bring my vision closer to reality. It’s exciting to explore new AI platforms knowing I will be able to share my journey to help others. I will continue to share updates once I build a timeline and roadmap. Here is what I am researching at this moment to help on this journey:

• Building a Notion learning plan with Google Gemini ‘Learn’ mode
• ChatGPT connectors, specifically GitHub and Notion
• Ghost API capabilities and limitations
• Zapier and n8n comparison
• Claude Skills platform

Automation didn’t launch this newsletter—discipline did. I’m sharing learnings in real time, and I haven’t mastered any of this yet. Three things keep me moving: learn the tool before automating it, build a stable weekly rhythm, and treat failures as scouting reports for the next experiment. I’m keeping costs low and myself in the editor’s chair while I test small, human-in-the-loop wins. Next week I’ll share another stumble—my early image and video-generation attempts—and how I’m working to improve. 

Goals & Milestones:

Follow us on social media and share Neural Gains Weekly with your network to help grow our community of ‘AI doers’. You can also contact me directly at admin@mindovermoney.ai or connect with me on LinkedIn.