- Back of the Envelope
- Posts
- ✉ Envelope #51: Size Effect Factor, AI Deep Research, and Hallucinations
✉ Envelope #51: Size Effect Factor, AI Deep Research, and Hallucinations
Andy Lin
May 09, 2025
Good morning! Andy from Back of the Envelope here.
If you've been using ACI 318-19, you may have come across the size effect modification factor, λₛ (lambda sub s), for nonprestressed members.
It reduces the concrete shear capacity quite a bit, so I wanted to dig into the "why." (And I’ll show you how I used AI to help me understand it better.)
Along the way, I also ran into a few AI hallucinations; so I’ll share why it’s so important to keep your critical thinking hat on when using AI for technical work.
Alright, let’s get into it!
(Estimated read time: 3 minutes)
The Code Change That Caught My Eye
Here’s what the equation for Vc looks like in ACI 318-19:
And the size effect modification factor:
This didn’t exist in earlier editions of ACI 318.
Basically, if your effective depth "d" is greater than 10 inches, and you’re not providing at least Av,min shear reinforcement, then your concrete shear capacity gets knocked down.
Here’s a quick table that shows how λₛ varies with d (the actual capacity is actually reduced more than that - I have another table coming up to show you):
At first glance, it feels like a minor annoyance: you may now need to add shear reinforcing to slabs that previously worked without it.
It’s kind of like the engineer’s version of, “I’ve been doing this for X years and it’s been fine!”
a reddit discussion several years ago when ACI 318-19 came out
Then the next thought hits: what if a savvy client notices the extra stirrups and questions it? How would I respond?
Cue the AI Deep Dive
That’s when I turned to Gemini Deep Research (2.5 Pro) and ChatGPT o3. (Gemini is free; ChatGPT o3 requires a Plus subscription.)
In Gemini, I selected "Deep Research" from the dropdown and typed:
In ACI 318-19, the shear capacity has dropped significantly with the size effect factor. This is the result of research findings that suggest concrete shear capacity does not increase linearly with deeper sections. Tell me everything you know.
It searched the internet and compiled a 24-page report that had basically everything (link to the report at the end).
I skimmed through it, and it turns out Professor Zdeněk P. Bažant at Northwestern had spotted this over 40 years ago. He built a theoretical framework to explain and quantify the phenomenon.
Over time, more tests confirmed his work, and eventually, it made its way into the code.
The idea is this: when shear cracks form, they get longer and open wider as they propagate down through deeper sections. That instability means the shear capacity per inch of depth actually gets worse with size.
But, If you add enough shear reinforcement (Av,min), the reduction doesn’t apply because cracks are effectively controlled.
AI also found a comparison table that helped conceptualize the effect (I didn’t double check the numbers so don’t take it for granted):
How to Explain It to Clients
Here’s the prompt I gave ChatGPT o3 (advanced reasoning model with internet search capability):
Hypothetically, I am designing a concrete structure nearly identical to the one I designed several years ago with the previous code cycle.
The current code references ACI 318-19 which has the size effect factor for shear, resulting in more shear reinforcing compare to the previous structure because concrete shear capacity is lower without reinforcing.
I anticipate the client (architect) and owner will potentially question the added reinforcing cost.
I understand this is based on research and the shear capacity doesn't increase linearly with depth, but how do I explain this without the client worry about their previous structure.
I am also now slightly concerned that the concrete structures designed over the past 20 years maybe under-designed.
Research and tell me your analysis and how I should explain this to the client in layman's terms. Keep the tone semi-casual and friendly. Avoid technical jargons.
After a few iterations (link to full chat at the end):
I had a decent draft of what I could tell the client (if this ever comes up)
Plus a summary of the relevant research
Here’s the gist:
Concrete sections without shear reinforcement still technically met the target probability-of-failure, but the safety margin was slim. The new factor is there to calibrate that margin.
In other words, it was safe (very little chance of failure), now it's safer (very very very little chance of failure).
A Quick Note on AI Hallucinations
This also brings up something we should all watch out for when using AI: hallucinations.
They can happen when AI takes a large volume of information, condenses it, and rephrases it. In the process, it can lose or distort the original meaning, context, or nuance.
(Honestly, the same thing happens with people too… we all simplify and reframe things, and sometimes, crucial information gets blurred or missed more than we realize.)
Anyhow, ChatGPT confidently told me:
"The research team behind λₛ looked at more than 4,000 test results and calibrated the new rule so that older, code-compliant members still meet a target reliability index."
It even linked to a PCI document.
When I checked the source? No mention of any reliability index.
When I pressed further, it eventually provided more context.
A important reminder: AI tools are fast and useful, but they’re not flawless. Always double-check.
That said, the time saved and clarity gained made it totally worth the effort!
Alrighty - that’s it for now. Thanks for reading!
If you enjoyed this one, pass it along to a loved one (or just a colleague who still keeps ACI 318-05 or 08 on their desk).
Best,
PS.
Working on a steel building? Check out Durafuse.
It’s a proprietary moment connection that offers high ductility and faster recovery after an earthquake.
Plus, they’ve got your back with design support: calcs, plan check comments, RFIs, shop drawing reviews, and a carwash — the whole package.
Click here to learn more: https://go.sehq.co/durafuse
(Jk about the carwash!)
PPS.
Links:
ChatGPT o3 discussion:
|
If you've made this far, might as well hit reply and let me know what you think of this week's email (because I love hearing from you!)
Oh, and tell others to sign up here.
Reply