What I Learned Running the Same Prompt Through 4 AIs 🤖

Spent time on the stationary bine today building an inflation model for Gen X retirement planning. Ran the identical prompt through Claude, ChatGPT, Gemini, and Kimi. Here's what I learned about how these models think differently.

The task: Build three inflation scenarios (base/bear/bull) for 2026-2045 with probability weights and multipliers showing what $100K today costs in 2040, 2045, 2060.

---

CLAUDE (my primary)

- Most conservative on Base Case at 3.2%

- Strong on structural reasoning: kept coming back to debt-to-GDP and Fed incentives

- Wanted to stay within "plausible bounds" - refused to model anything above 7% or below 1%

- Best at taking inputs from the other three models and synthesizing them into a coherent framework

- When I pushed back on assumptions, it adjusted thoughtfully rather than just agreeing

CHATGPT

- Most optimistic overall: Base Case only 2.4%, expected rate 2.64%

- Anchored heavily to Fed targets and professional forecaster surveys

- Provided the most citations and links to primary sources (Fed statements, BLS, CBO)

- Clean methodology explanation with all the math shown step by step

- Best at explaining the "official" view of how inflation is measured and why CPI works the way it does

GEMINI

- Most pessimistic: Bear Case at 5.3%, expected rate 3.47%

- Introduced "fiscal dominance" framing that became central to our final model

- Best articulation of why the downside risks are structural, not hypothetical

- Bear Case 2060 multiplier of 5.86x was the scariest number anyone produced

- Really good at stress-testing assumptions and explaining what could go wrong

KIMI (via Maple AI)

- Middle of the pack on the numbers themselves

- Introduced the "personal inflation premium" concept: add 1.75% to CPI for experienced inflation

- This was the most practical innovation of the day - it acknowledges the CPI gap without requiring us to build alternative indices

- Good at framing outputs for actual retirement planning use, not just academic analysis

---

WHERE THEY AGREED 🤝

- All used 50-55% for Base Case probability

- All put Bull Case at 20% (nobody believes the productivity miracle is likely)

- All structured piecewise rates (near-term higher, long-term normalizing)

- All acknowledged CPI understates lived experience but used it anyway for consistency

WHERE THEY DIVERGED 📊

Base Case rates:

- ChatGPT: 2.4% (optimistic)

- Kimi: 2.8%

- Gemini: 3.0%

- Claude: 3.2% (conservative)

Bear Case rates:

- ChatGPT: 4.0%

- Kimi: 4.5%

- Claude: 4.8%

- Gemini: 5.3% (most aggressive)

The spread on expected 2060 income needed (to match $100K today):

- ChatGPT: $248K

- Kimi: $291K

- Claude: $352K

- Gemini: $352K

---

WHAT I TOOK FROM THIS 🎯

1. Each AI has a personality, and it shows up in the numbers. ChatGPT trusts institutions and anchors to official forecasts.

Gemini is skeptical and looks for structural risks. Kimi focuses on practical application. Claude tries to find the middle ground. None of them is "wrong" - they're just approaching the same problem from different angles. Running all four gave me a much better sense of the reasonable range than any single model would have.

2. ChatGPT is best when you want the mainstream consensus view.

It cited Fed statements, CBO projections, and professional forecaster surveys. If you need to understand what the official position is, or you want your work to be defensible against institutional criticism, ChatGPT will give you that. The tradeoff is it may be too anchored to what authorities say rather than what's structurally likely.

3. Gemini is the one to use when you want your assumptions stress-tested.

It kept finding reasons why things could go wrong. The "fiscal dominance" framing - where the Fed loses independence because the government can't afford higher rates - was Gemini's contribution, and it became central to our Bear Case. If you're building something important, run it through Gemini and ask "what am I missing?"

4. Kimi surprised me with practical innovation.

The "personal inflation premium" idea - just add 1.75% to CPI rates for your personal planning - was simple and useful. It solved the whole "CPI doesn't match my experience" debate without requiring us to build alternative indices. Sometimes the best insights are frameworks, not numbers.

5. Claude is good at synthesis but needs to be pushed.

My initial Bear Case was 4.8%, which felt too tame after seeing Gemini's 5.3%. When I pushed back, Claude adjusted to 5.0% and explained why. The lesson: don't just accept the first output. The models get better when you engage with them critically.

6. The disagreement itself is valuable information.

When ChatGPT says 2.4% and Gemini says 3.0% for the Base Case, that 60 basis point spread tells you something about the uncertainty in the system. The "right" answer isn't splitting the difference - it's understanding why smart systems disagree and making a judgment call about which reasoning you find more compelling.

7. Probability weights matter more than point estimates.

All four models agreed that the Bear Case deserves 25-30% probability, not 10%. That's nearly one-in-three odds of a bad outcome. When you're planning for a 35-year retirement horizon, one-in-three odds of needing $550K instead of $300K to match today's $100K is not something you can ignore. The expected value masks this risk.

8. None of them pushed back on the premise.

I asked for inflation scenarios, and all four delivered inflation scenarios. None of them said "wait, maybe you should also consider deflation" or "have you thought about currency regime change?" The models are good at answering the question you ask, but they're not great at questioning whether you're asking the right question. That's still your job.

9. Running multiple models takes more time but builds more confidence.

This whole exercise took about 50 minutes of prompting, comparing, and synthesizing. I could have just used Claude and been done in 15 minutes. But I wouldn't have discovered the fiscal dominance framing, or the personal inflation premium concept, or understood why the mainstream view might be too optimistic. The extra time was worth it for a model I'll use repeatedly.

10. The final output is better than any single model produced.

Our consensus landed at 3.46% expected inflation with a 50/30/20 probability split. That's more conservative than ChatGPT, less alarmist than Gemini, and incorporates the best ideas from all four. The whole really is greater than the sum of the parts - but only if you do the synthesis work yourself.

---

THE CONSENSUS MODEL WE BUILT

After comparing all four, landed here:

Base Case (50%): 3.2% average

Bear Case (30%): 5.0% average

Bull Case (20%): 2.0% average

Expected: 3.46%

To maintain $100K purchasing power:

- 2040 (retirement at 65): $171K needed

- 2045 (age 70): $202K needed

- 2060 (age 85): $352K needed

The Bear Case is the wake-up call:

If things go wrong, you need $552K in 2060 to match what $100K buys today. That's 5.5x. Traditional retirement planning doesn't account for this.

---

Next step: plug these multipliers into a Bitcoin accumulation calculator. The question becomes: how much BTC does a Gen Xer need to clear the 3.5% real return hurdle for 35 years?

That's tomorrow's project. 😀👊🚴

#genx #inflation #bitcoin #retirement #ai