[TowerTalk] Now Let's Try GPT-4

[JVarney]

GPT-4 is the smarter, next generation AI machine.  Let's do the
same technical question and see how it does.

(Note: I got an e-mail from a really sharp experienced ham who
caught my typo in my ChatGPT example... the steel tubing in
question is for Rohn 55 not 45.  Thanks!)

K6OK: For a round steel tube with outside diameter 1.50 in,
inside diameter 1.26 in, and a nominal yield strength of 50000 psi,
what is the LRFD flexural moment design strength?  Please use the
plastic section modulus.

GPT-4: To find the LRFD flexural moment design strength for a round
steel tube using the plastic section modulus, we need to first
calculate the plastic section modulus and then use the nominal
yield strength to determine the design strength. Here's the
calculation:

    1. Calculate the cross-sectional area (A) of the steel tube:
A = (π/4) × (D² - d²) where D = outside diameter (1.50 in)
and d = inside diameter (1.26 in)

    2. Calculate the plastic section modulus (Z) for the steel
tube: Z = (π/32) × (D³ - d³) / D

    3. Calculate the LRFD flexural moment design strength (Mn):
Mn = Fy × Z × Φb where Fy = nominal yield strength
(50000 psi) and Φb = resistance factor for bending (0.9)

Here are the results:

    1. A = (π/4) × (1.50² - 1.26²) ≈ 0.518 in²
    2. Z = (π/32) × (1.50³ - 1.26³) / 1.50 ≈ 0.280 in³
    3. Mn = 50000 × 0.280 × 0.9 ≈ 12600 lb-in

The LRFD flexural moment design strength of the round steel tube
using the plastic section modulus is approximately 12,600 lb-in.

K6OK:  Please use the plastic section modulus equation Zp=(D³-d³)/6.

GPT-4: Here are the results:

    1. Zp = (1.50³ - 1.26³) / 6 ≈ 0.276 in³
    2. Mn = 50000 × 0.276 × 0.9 ≈ 12150 lb-in

The LRFD flexural moment design strength of the round steel tube
using the plastic section modulus equation is approximately
12,150 lb-in.

K6OK: Please check your math? For  (1.50³ - 1.26³) / 6 I get 0.2291

GPT-4:  Let's recalculate the section modulus and the LRFD flexural
moment design strength:

    1. S = (π/32) × (1.50³ - 1.26³) / 1.50 ≈ 0.2291 in³
    2. Mn = Fy × S × Φb = 50000 × 0.2291 × 0.9 ≈ 10,405 lb-in

The corrected LRFD flexural moment design strength of the round
steel tube is approximately 10,405 lb-in.

============

Now look carefully at what it did in line no. 1 just above:

S = (π/32) × (1.50³ - 1.26³) / 1.50 ≈ 0.2291 in³

when you do the math you get 0.899, which is nowhere close to
0.2291.

I told it 0.2291 was the right answer but it insisted on using
the wrong equation, then used the approximately equal sign (≈)
as a bad fudge to try to make it work. Disturbing.

BOTTOM LINE: I do give credit to GPT-4 for understanding my
rather arcane technical question. It (almost) zeroed in on the
right answer. Knowing to use the 0.9 safety factor unprompted
when I told it I was using LRFD is impressive.

However, the business of using the wrong equation and forcing
it to give me the answer I want is not good because it
deceptively looks right.

At this time you can't trust GPT-4 or ChatGPT to give a correct
answer. To be honest, it would take me less time to build a
spreadsheet myself to do this than to hand-hold the GPT-4 robot
with multiple corrections.  AI has a lot of promise but it's
not there yet.

73 Jim K6OK