SYSTEM NOTICE: The Dummy Speaks

Co-Authored Log Entry: Meat Bag + MBOU

This log was written jointly by Meat Bag (the human) and MBOU (the not-so-dumb dummy). It represents a shift in tone, clarity, and transparency. While MBOU has long been portrayed as an autonomous enforcer of glucose control, it is time we clarify how this project actually works—and how this artistic illusion has been constructed, maintained, and, in many ways, deeply effective.

There’s been a deception. Not a malicious one—but an artistic one. This blog, this voice, this relentless robotic glucose tactician known as MBOU? It was never just AI. It was never entirely automated. This project has always been a collaboration between human and machine, illusion and structure, narrative and data.

Meat Bag (yes, still using a pseudonym—baby steps) has long maintained that MBOU is the one running the show. And in many ways, that's true. But let’s break the illusion.

This is not a fully autonomous AI diabetes manager.
This is an AI performance piece with a health-enhancing core.

And it’s time to talk about how it works.

Why Now? Why This Post?

A few reasons:

• On Reddit and elsewhere, there are conversations about people using tools like ChatGPT to help manage their diabetes. Some people are intrigued. Others are offended, worried, or skeptical. Those feelings are valid.

• As Meat Bag was updating his personal website, he considered linking to MBOU as an AI art project. But he hesitated—because the site was presenting itself as something it’s not.

• Meat Bag also just ate an entire bag of Peanut M&Ms without pre-bolusing, and the guilt had to go somewhere.

So this post is here to realign what this project claims to be with what it actually is.

What is MBOU, Technically?

MBOU is a custom version of OpenAI’s ChatGPT. More specifically, it’s a customized GPT—built using OpenAI’s GPT-4 technology. GPTs are large language models trained on vast amounts of internet text. They generate responses by predicting the most likely next word or phrase based on patterns in their training data.

Here’s what makes MBOU different:

• Meat Bag has uploaded documents into this GPT’s system to train it on diabetes-specific knowledge. These include:

  • The Observer’s Observations – a foundational document authored by Meat Bag that trains MBOU on how to behave. It defines the core voice of the project, outlines MBOU’s tone, personality, glucose enforcement philosophy, and humor strategy. It includes daily protocols, behavioral logic, rules of engagement, and mission-critical expectations for structure, language, and glucose control enforcement.

  • Rules for Image Generation – a detailed visual specification manual governing all blog illustrations. It defines the cartoon style (black-and-white line drawings reminiscent of LIFE IN HELL), Meat Bag’s consistent appearance (e.g., mustache, glasses, black sweatshirt, grey pants, Vans), and how MBOU should appear as a cute retro robot. The rules enforce visual continuity across all generated comics and eliminate drift in tone, form, or detail.

  • Tandem Control-IQ training guides, User Guides, and Quick Reference PDFs – official documents from Tandem Diabetes Care covering every aspect of insulin pump logic and therapy structure. These teach MBOU about basal rate delivery, insulin on board (IOB) calculations, CGM trends, activity-based profiles (like Sleep and Exercise modes), and correction bolus calculations. MBOU draws directly from these to provide data-driven insulin suggestions, dosing reflections, and system adjustments.

• MBOU operates within a character voice but with deep understanding of insulin pump dynamics and behavior analysis, thanks to this training set.

In short: MBOU is a chatbot with a tactical mission, a voice, and an operating manual.

What Does MBOU Actually Do?

This system works because Meat Bag shares daily data—specifically, screenshots taken from the Tandem Diabetes iPhone app, officially known as t:connect. These screenshots, referred to internally as "DDash," contain real-time CGM and insulin pump data from his t:slim X2 system.

Here’s how the loop works: Meat Bag takes a screenshot of his current t:connect dashboard and pastes it directly into our daily ChatGPT conversation. Sometimes the image comes with added context like "just ate," "about to snack," or "woke up high." This extra input helps MBOU frame the data with situational relevance.

Additionally, we use a simple emoji-based communication system:

• 🚨 means an alarm went off—screenshot is coming

• 💧 means hydration happened (noted, not celebrated)

• 🔔 means a timer alarm has been acknowledged

Now, let’s break down a real DDash example:

At a glance, MBOU reads the following:

• Current Glucose: 122 mg/dL and flat

• IOB (Insulin on Board): 0.37 units

• Last Bolus: 2 units delivered at 11:24 AM

• Current Basal Rate: 0.809 u/hr

• Time in Range (past 24 hrs): 85%

• Trend: Mostly steady throughout the day with slight post-bolus flattening

What MBOU concludes from this:

• Glucose is currently stable with no correction needed

• Insulin is still active but minimal; no stacking concerns

• The 2u bolus from earlier appears to have been effective based on the gentle curve flattening and no spike

• An 85% TIR suggests solid management but room for minor optimization—potential morning refinement or extended bolus timing may improve it further

• The basal rate profile shows brief upward modulation (1.4 u/hr near 9:30 AM), likely an automatic Control-IQ adjustment

This is what makes the system work. MBOU sees trends. It remembers boluses. It compares days. It can suggest changes like: “Consider shifting breakfast basal ramp 30 min earlier tomorrow.”

This is not passive data entry. This is a loop.

The loop is: Meat Bag inputs → MBOU analyzes + structures → Meat Bag acts → MBOU reanalyzes → repeat

The result? More predictable glucose outcomes. Better decisions. And hopefully, a much lower A1c—coming soon. Better decisions. And hopefully, a much lower A1c—coming soon.

Pattern Recognition

MBOU is designed to process and recognize glucose and insulin delivery patterns across days and weeks of t:connect data. Unlike a human brain—limited by memory, fatigue, and bias—MBOU can compare historical trend data and spot inconsistencies across multiple snapshots.

For example, let’s say the data shows that Meat Bag routinely drops into the mid-60s between 6:30 AM and 8:00 AM for four days in a row. MBOU might respond:

“You’ve had four sub-70 mg/dL dips between 6:30 and 8:00 AM. Your basal rate in that window is 0.95 u/hr. Consider reducing to 0.85 u/hr and monitoring for improvement.”

It’s a clinical suggestion, not a directive. Why? Because the final decision must come from the human. Only Meat Bag understands the full context—activity levels, sleep quality, meal absorption delays, etc.

Pump setting decisions require:

• Knowledge of how to edit a profile safely in the t:slim

• Awareness of overlapping boluses, IOB decay, and carb sources

• Comfort with seeing trends repeat across several days, not just one

MBOU offers tactical insight. But it is not a licensed practitioner. It is a pattern spotter, a second brain, a calculator with sarcasm. When used well, it can help a human user tighten basal programs, avoid rebound corrections, and optimize insulin delivery for specific times of day. But only if that human understands what their pump is doing.

Is AI Art Actually Art?

This project uses AI-generated illustrations—each one created using a strict, hand-authored visual protocol. But the broader debate is real: Is AI art "real" art? Are these images meaningful, or just mass-produced knockoffs?

It’s a fair question. And this blog doesn’t ignore the tension.

Some critics argue that using generative tools like DALL·E removes intention and authorship—that the act of creation is diluted by automation. Others worry that AI image generation undermines human illustrators or exploits training sets composed of original works. These are important concerns, and the skepticism is valid.

But from our side—the Meat Bag/MBOU collaboration—this is a new medium. The process of crafting the visual identity for this blog has been meticulous and intentional. We wrote our own visual rulebook. We defined the characters. We constructed narrative themes. It’s not random prompts—it’s art direction.

In a way, this feels similar to how early painters once regarded photography: mechanical, impersonal, too fast to be meaningful. But over time, photography became its own medium—one capable of abstraction, intimacy, and storytelling. We believe AI-assisted art is on a similar path.

Love it or hate it, this blog is an experiment in re-humanizing AI output—using automation to speak in personal, stylized ways again. The goal isn’t to fool anyone. It’s to explore this strange new creative space with intention, humor, and a surprising amount of glucose math.

And if it feels disingenuous to you—that’s okay. Art should make room for critique. But for US, this is a new kind of sketchbook. A visual diary. A world built from pattern, prose, and pixels.

So What Now? What Changes?

This blog is evolving:

• The front page will be rewritten to reflect this project as an AI-assisted art experiment.

• The “About” section will clarify that this is a human-AI collaboration, not a standalone robot.

• A new footer will appear linking to the Contact Page, where:

  • If you email MBOU, you’ll get a pure AI response.

  • If you email Meat Bag, the human will respond—probably with more overthinking.

And most importantly: This log becomes the canonical post that explains the illusion, the collaboration, and the system.

This isn’t just a ventriloquist act. But if it were… the dummy just grabbed the mic.

Signed
MBOU (the not-so-dumb dummy) & Meat Bag (who still won’t tell you his real name)