Hacker News new | past | comments | ask | show | jobs | submit login

OK I tried something like my favourite question: "What are the most important factors in a simply supported beams's resistance to bending and shear strength?"

That prompt is deliberately a bit humanly stated whilst being quite obvious for any first year Struct/Civ eng student who actually made it into a text book. It is far simpler than any programming prompt effort I've ever tried with say ChatGPT.

The answer I got was: "In conclusion, the most important factors in a simply supported beam's resistance to bending and shear strength are the forces and moments acting on the beam."

I changed my prompt to: "What are the most important factors in a simply supported beams's resistance to a point load?"

"In conclusion, the resistance of a simply supported beam to a point load can be affected by several factors, including the reaction forces and moments, the second moment of area and shear coefficient, and the material properties of the beam. There are multiple approaches to calculating the resistance, including hand calculations and finite element analysis, each with their own pros and cons."

Much better but largely rubbish. Material properties is correct.

Think about a steel I beam - why is it mostly air in cross section? You even see I beams where the web (the vertical bit) has holes in it (to reduce weight). The resistance of a simply supported beam in response to a point vertical load is purely down to vertical depth of the member - it's a classic result from quite a lot of engineering math at 1st year uni. The flanges ie the horizontal top and bottom bits are to resist buckling and shear. Yes, wood gets a bit more complicated and actually all materials get more complicated!

However, I find that LLMs (like I've seen loads!) seem to have snags with non IT related engineering stuff. ChatGPT seems to have intimate knowledge of VMware PowerCLI and Python (and eventually wrote me a decent script) but fails on a basic physical principle.

Caveat Civis




Did you use Expert mode? With Expert mode I got an answer that seems correct: https://www.phind.com/search?cache=43c579b1-6b5f-4665-94e4-a...


That answer is quite close but probably by accident. Don't get me wrong, I am in absolute awe of these beasts:

Me: "Take a simply supported beam 10m long. Apply a force of 1000N at 2m from the left of that beam. What is the bending moment at 5m from the left?"

It: "blah blah blah ... Therefore, the bending moment at a point 5 meters from the left end of the beam is 600 Nm."

It's probably correct here but I will have to check. I've just read its reasoning and it does look correct on a superficial reading after two very large glasses of wine. Sadly it fails to note that I spelled metre in the French way - obviously, because I am English.

This is the result:

https://www.phind.com/search?cache=b2d2e3ee-f3ed-4a2a-b976-5...


Change the name of expert mode. You're having to tell each single user to turn it on, because probably like me they think it means "answer as if talking to an expert"




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: