Hey everyone! 👋

We are starting December in the middle of an AI upgrade wave: new models, more powerful hardware, and national projects treating AI like core infrastructure.

In AI Education, we kick off a four-part large language model deep dive so you finally have a clear mental model for what systems like ChatGPT are actually doing under the hood. This week’s 10-Minute Win turns scattered debts into a simple Snowball or Savvy paydown gameplan you can stick with. And in Founder’s Corner, I share honest reviews of the latest models and how they perform so you can pick the right one for whatever task you tackle next.

1) Claude Opus 4.5 is here

Summary: Anthropic launched Claude Opus 4.5, calling it its best model yet for coding, agents, and computer use, with big upgrades in deep research, spreadsheets/slides work, and long-running workflows.

Why it matters: This is another clear step into the agent era—models that don’t just chat, but reliably operate tools, automate multi-step tasks, and keep context over time. That’s where real productivity (and enterprise spend) lives.

2) AI helps drive record $11.8 billion in U.S. Black Friday online spending

Summary: U.S. shoppers spent a record $11.8B online on Black Friday, up 9.1% year-on-year, with Adobe Analytics citing an 805% surge in AI-assisted traffic as tools like Walmart’s Sparky and Amazon’s Rufus steered people to deals.

Why it matters: This is AI moving real consumer dollars. Retail copilots and recommendation agents aren’t side experiments—they’re now measurable drivers of sales and a preview of how AI will influence everyday spending.

3) Amazon to invest $50bn in AI for US government customers

Summary: Amazon plans to invest up to $50B to expand AI and supercomputing capacity for U.S. government customers, building out what’s effectively a massive, government-focused AI cloud under AWS.

Why it matters: This is AI infrastructure as national capability. It locks AWS even deeper into defense, intel, and public-sector workloads—and signals long-term, contract-backed demand for AI compute.

4) Alphabet on pace to hit $4 trillion market value as AI gains momentum

Summary: Alphabet is closing in on a $4T valuation after a year-long rally driven by its AI pivot—Gemini 3, TPU strategy, and cloud growth—sending shares up ~4% in recent trading.

Why it matters: Google joining the multi-trillion race on the back of AI underlines a simple point: markets now treat AI as core infrastructure, not a bolt-on feature set.

5) Launching the Genesis Mission

Summary: A new Executive Order launches the Genesis Mission, a national effort to use AI and U.S. supercomputers to accelerate scientific discovery, coordinate federal labs, and tackle “grand challenge” research problems.

Why it matters: This is the U.S. leaning into AI-as-R&D-engine—treating AI like a Manhattan/Apollo-style capability for science. It sets direction (and future grant money) for labs, universities, and companies working at the AI–science frontier.

Over the past few volumes, you have picked up the building blocks:

• Artificial intelligence as the broad idea
• Machine learning as systems that learn from data
• Deep learning and neural networks
• Tokens, vectors, embeddings, prompts, and context windows

This week we put a clean frame around all of that: What is a large language model really?

Not the marketing answer. The mental model you can hold in your head when you open an app like ChatGPT and start typing.

By the end of this four-part series, you should feel like you actually know what is happening under the hood when you ask AI to explain a bill, summarize an email, or help you think through a big financial decision.

What a large language model actually is

A large language model is:

A system that has learned patterns in text and uses those patterns to predict the next small piece of text, over and over.

That is it at the core. It does not “understand” the world like a human. It is extremely good at continuing text in ways that match patterns it has seen during training.

When you see a fluent answer, you are seeing:

• Many small predictions chained together
• Guided by patterns the model learned from huge amounts of text
• Shaped by the prompt and context you give it

How it is different from classic software

Classic software:

• Follows hand-written rules
• If X happens, do Y
• Every rule is coded by a developer

A large language model:

• Is not given explicit rules for everything
• Learns patterns from data
• Generalizes to new prompts it has never seen before

Analogy: Classic software is like a set of tax forms with instructions. A large language model is like a person who has seen millions of completed forms and can guess how you should fill yours out, just by pattern.

The two phases: training and use

There are two big phases in the life of a model.

1. Training phase
   • The model sees huge amounts of text
   • It learns which tokens tend to follow which other tokens
   • Its internal settings are adjusted over and over to reduce mistakes
2. Use phase (when you chat)
   • The training is frozen
   • You send a prompt
   • The model turns your text into tokens and predicts the next ones
   • It uses what it learned during training to shape its answer

Another way to say it:

• Training is where the model learns patterns
• Use is where the model applies those patterns to your prompt

Why it feels so smart

If you train a system on huge amounts of:

• Explanations
• Arguments
• Instructions
• Code
• Articles and conversations

And you force it to practice predicting the next token billions of times, you get behavior that looks like reasoning:

• It can follow steps
• It can compare options
• It can explain concepts in different ways

But under the hood it is still doing one basic thing:

Predicting the next piece of text that would make sense here, given everything it has seen before and the input you just gave it.

FAQs

Q: If it only predicts the next piece of text, how can it solve real problems?

A: Because many real problems show up in text form: questions, emails, documents, code, instructions. The patterns it learned cover not just words, but also how humans explain, reason, and structure answers. Predicting the next piece of text over and over lets those patterns show up in useful ways.

Q: Is a large language model the same as artificial intelligence?

A: No. It is one kind of AI system. It is a very powerful one for language, but AI also includes other models and approaches that handle images, audio, planning, and more.

Closing this week

This week you got the high-level mental model:

• A large language model is a pattern learner for text
• It learns in a training phase and applies those patterns when you chat
• It feels smart because language carries a lot of human reasoning patterns

Next week we zoom into how it learns:

• What the training data looks like
• How the model adjusts its internal settings
• Why it needs so much data and compute
• How data quality shapes its strengths and blind spots

Think of Part 1 as “what this thing is”. Part 2 will be “how it actually learns”.

💳 Debt Paydown Gameplan

Why this matters: Carrying multiple debts is like running with a weighted vest — you can move, but everything is harder. The problem: most people either avoid looking at the full picture or randomly throw extra money at whatever hurts most that month. In 10 minutes, this workflow will help you list every debt in one place, compare the classic Snowball (smallest balance first) vs. Savvy (highest interest first) approaches, and walk away with a simple, AI-assisted paydown plan you can actually follow.

Step 1 — List your debts in one simple sheet (2–3 minutes)

Open Google Sheets and create this table:

Then replace the sample numbers with your debts:

• Include: credit cards, personal loans, BNPL, auto loans, anything with a monthly payment.
• Exclude for now: mortgage and student loans if you want to keep it focused (you can run a second pass just for those).

Below the table, add one more cell:

• In A7, type Extra_Payment_$.
• In B7, enter how much extra you could realistically send toward debt each month (even $25–$50 helps).

If you already used earlier 10-Minute Wins (budget or net worth), you can copy balances from those sheets straight into this table so everything stays consistent.

Step 2 — Get AI to build Snowball vs. Savvy plans (4 minutes)

Copy your entire table (including headers and the Extra_Payment_$ cell) and paste it into a new ChatGPT 5.1 chat with this prompt:

Role: You are my debt paydown planner.

Context: I want a clear, simple plan to pay off my debts using both the Snowball method (smallest balance first) and a more Savvy method (highest interest rate first), so I can choose which fits me better.

Data: Here is my debt table with columns: Debt Name, Balance, APR, Minimum Payment, Type, plus a row showing how much extra I can pay each month.

Instructions:

1. Verify the table and restate my total debt and total minimum payments.
2. Build two payoff orders:
   • Snowball: sort by smallest Balance first.
   • Savvy: sort by highest APR first.
3. For each method, assume I pay all minimums and send my full Extra_Payment_$ to the current focus debt until it’s gone, then roll that amount into the next one.
4. For each method, output:
   • A table: Order | Debt Name | Focus or Minimum | Approx. Months to Clear Each Debt | Approx. Total Months to Debt-Free (rough estimates are fine).
   • 3 bullet points on pros/cons for me (behavioral + mathematical).
5. End with a 1-paragraph recommendation on which method might fit me better (you can’t give financial advice, but you can talk through tradeoffs).

Rules:

• If the numbers look impossible with my current Extra_Payment_$, say so gently and suggest starting with a smaller goal (e.g., first debt only).
• Keep the language plain and encouraging—no jargon.

You’ll now have two clear roadmaps with a recommendation.

Step 3 — Pick your method and set your “focus debt” (2 minutes)

Back in Google Sheets:

1. Add a new column F (Strategy_Focus) and mark, for the current month, which debt is your focus under your chosen strategy (Snowball or Savvy).
   • Example: put Snowball_Focus next to the smallest balance, or Savvy_Focus next to the highest APR.
2. In your notes (or a new cell), write:
   • My method: Snowball or My method: Savvy
   • Current focus debt: [Debt Name]
   • Extra I’m sending to this debt each month: $[Extra_Payment_$]

Step 4 — Turn the plan into a monthly habit (2–3 minutes)

• Save your plan: paste ChatGPT’s summary (Snowball vs. Savvy comparison and your chosen method) into a Google Doc or Notion page titled: Debt Paydown Gameplan — [Month Year].
• Set a reminder: in Google Calendar, create a monthly event on or right after your pay date:
  • Title: Send $[Extra_Payment_$] to [Focus Debt Name].
  • Description: paste the link to your sheet/doc plus your chosen method (Snowball/Savvy).
• Every 1–3 months, come back to this same workflow: update balances, re-run the prompt, and move the Focus tag to the next debt when one is paid off.

The Payoff

Instead of “I’ll just throw extra at something,” you now have:

• Every debt in one clean list
• Two structured payoff orders (Snowball vs. Savvy) generated for you
• A single focus debt and extra payment amount for this month
• A recurring reminder that turns your decision into a habit

You’re not guessing—you’re executing a simple, chosen gameplan.

Transparency & Notes for Readers

• All tools are free: Google Sheets, ChatGPT Free, Google Docs/Notion, Google Calendar; the CFPB calculator is optional and free.
• Estimates only: The timelines are rough—interest accrues daily and payments can change. Always check with your lender for exact payoff details.
• Risk: Don’t stop making minimum payments on any debt while using this.
• Educational workflow — not financial advice.

We’ve officially entered the holiday shopping season where companies compete for our attention and our money. It seems like the AI companies are tapping into our consumer habits with the flurry of announcements over the last few weeks. Google, OpenAI, and Anthropic have all dropped new models that are changing the game and bringing new intelligence into their tools. It’s overwhelming for anyone following the AI release cycle, but these announcements are increasingly important for us to track and understand. The tech is moving fast and we risk losing productivity gains, at home and at work, by falling behind. This week, I am focusing on three of the latest: Gemini 3 Pro, ChatGPT 5.1 and Claude Opus 4.5, and sharing how I use these models on a daily basis.

ChatGPT 5.1

From Day 1, I’ve relied heavily on the latest ChatGPT model to be my business partner and content creator. One major reason is the ‘Projects’ feature, which allows me to capture all of the context from my project in one area. This trend has continued since GPT 5.1 was released in November. I personally like the model’s ability to follow instructions and consistently deliver outputs that stay within the parameters I’ve outlined. GPT 5.1 excels at reasoning and putting together executable plans in an easily digestible manner, which is extremely useful when tackling problems in a domain outside of my personal expertise. Writing is a main strength of this model, especially when you build up memory context within a project. The model can pull from previously approved writing outputs and use it as a template to mimic the style and tone. This helps create a consistent ‘reading feel’ in each weekly newsletter. One suggestion I have is to experiment with ‘5.1 Thinking’ and track the ‘thought’ process of the model. It’s impressive to see how the model works through multiple steps to formulate comprehensive responses. It feels human-like and is the main reason I haven’t fully pivoted to another model.

Pro Tip: Use the ‘Projects’ feature to house and organize a project, whether it be personal or work related. 

Gemini 3.0

I’m definitely most experienced with the suite of ChatGPT models, but Gemini 3.0 is slowly poaching scope and tasks in my everyday workflows. Google has been crushing it with their models ever since Gemini 1.5 was released in February 2024. Historically, I’ve used Gemini 2.5 for two main tasks: helping write LinkedIn posts to promote the newsletter and creating images with the Nanobanana image model inside Gemini. This lack of usage was mostly due to my comfort and project integration with ChatGPT. But this is changing due to Gemini 3.0. If I had to describe 3.0 in one word, I’d go with ‘Powerhouse’. Google has figured out how to wrap their various models into one experience. For example, you can prompt Gemini 3.0 to create images, which is powered by their Nanobanana model. Or you can create realistic videos by tapping into Veo 3. Gemini 3.0 excels in multimodality, allowing the consumer to seamlessly interact with one interface to accomplish any task, regardless of the complexity. Personally, I started using 3.0 to help me build structure in the ‘Founder’s Corner’ section of the newsletter. I found myself wasting time with writer’s block and needed to prioritize planning over execution. The model blew me away with the level of detail given to help me organize my thoughts. I used this process for this blog post and probably saved myself at least an hour of writing time, all thanks to the power of 3.0. I’m still experimenting with 3.0 (it’s only been out for two weeks) and I have no doubt that more of my work will shift over once I get more reps under my belt. 

Pro Tip: Use Gemini 3.0 in Google AI Studio to vibe code a simple app that can make your life easier. You will get a chance to see the model use reasoning to bring your idea to life.

Opus 4.5

It’s hard to experiment with every model. I pay for ChatGPT and Gemini (the $20 per month plans) and have not taken the leap to pay for Anthropic’s suite of Claude models. One reason is how Claude is positioned. These models are trained and built to excel at coding and are marketed primarily toward developers. I don’t have advanced use cases (yet) that justify shifting over to Claude models, but Opus 4.5 might force me to rethink that strategy. Since I don’t have direct experience with this model, I asked Claude Sonnet 4.5 to highlight the strengths of Opus 4.5 and tell me why I should switch. Here are the main bullets/selling points:

• World's Best Coding Model
• Built for Agentic Workflows
• Computer Use & Browser Automation Leader
• Enterprise & Office Productivity Powerhouse
• Unprecedented Efficiency

My goal is to start incorporating Claude models into projects as I build out my 2026 roadmap. There are projects that fit nicely with the selling points and strengths of Opus 4.5. It’s good to step outside of your comfort zone, especially with the newer AI models.

Pro Tip: Find the right use case before experimenting with a new model, especially if you have little to no experience with previous versions. 

I view each model release as an opportunity to learn new skills and keep up with the pace of change. Each update provides a unique chance to figure out the best prompt or data structure that allows the model to create your envisioned output. This matters at work, because the tools you use every day may quietly swap the underlying model, and the inputs that used to perform well can suddenly fall flat if you do not understand how that model behaves. The more familiar you are with Gemini, ChatGPT, and Claude, the easier it is to adapt when your company or your favorite app changes what is running under the hood. Next week, I will build on this and share three practical ways to experiment with these models so you can accelerate your learning and get more out of AI in both your work and your everyday life.

Goals & Milestones: