AI fails

[Ian H]

Wanting to replace a hub gear, I googled "Sachs Torpedo flange diameter". I got this:-
The Sachs Torpedo "flame pipe" (exhaust manifold) diameter varies depending on the specific engine model, but common sizes are 28 mm and 32 mm.

 

[briantrumpet]

It seems that ChatGPT is finding it hard to keep up with the actualité

 

[briantrumpet]

Wanting to replace a hub gear, I googled "Sachs Torpedo flange diameter". I got this:-
The Sachs Torpedo "flame pipe" (exhaust manifold) diameter varies depending on the specific engine model, but common sizes are 28 mm and 32 mm.

I guess it got bored after "Sachs Torpedo", dozed for a bit, then got excited by "diameter". Could happen to any of us.

 

[icowden]

Wanting to replace a hub gear, I googled "Sachs Torpedo flange diameter". I got this:-
The Sachs Torpedo "flame pipe" (exhaust manifold) diameter varies depending on the specific engine model, but common sizes are 28 mm and 32 mm.

If you ask Copilot (powered by Chat-GPT Smart GPT-5) "tell me about Sachs Torpedo flange diameter" it does seem to know what it is talking about. This suggests that Chat-GPT is much more advanced than Gemini. I've found the same with coding. Copilot does better than Gemini:

Sachs Torpedo Hub Flange Details​

The term flange diameter refers to the size of the circular flange on the hub shell where the spoke holes are drilled. This measurement is important for wheel building, as it affects spoke length calculations and wheel strength.

• Sachs Torpedo 3V steel silver
  • Flange diameter: 68 mm
  • Flange distance: 47 mm
  • Spoke holes: 28
  • Axle length: 145 mm
  • Weight: 915 g
• Sachs Torpedo Pentasport H5111
  • Early models had flanges machined from the same steel piece as the hub shell, giving them a robust and corrosion‑resistant chrome finish.
  • Later versions switched to pressed‑on stamped sheet metal flanges, which reduced machining time and material use.
• Fichtel & Sachs Torpedo 53
  • Produced around 1955.
  • Detailed specs include rear spacing (115 mm) and spoke drilling (36 holes).
  • Flange size is listed but varies depending on production year.

⚙️ Why Flange Diameter Matters​

• Spoke length calculation: Larger flange diameters shorten spoke length, while smaller ones lengthen it.
• Wheel strength: Bigger flanges can improve lateral stiffness and distribute stress more evenly.
• Compatibility: Different Torpedo models (3‑speed, 5‑speed Pentasport, coaster brake versions) may require different spoke lengths and lacing patterns.

📌 Summary​

If you’re working with a Sachs Torpedo 3‑speed hub, expect a flange diameter of 68 mm. Other models, like the Pentasport or older Torpedo 53, may differ depending on design era and brake type. Always measure directly or consult the specific hub’s technical sheet when building wheels.

Would you like me to create a comparison chart of flange diameters across different Sachs Torpedo models or focus on how to calculate spoke lengths using flange diameter?

 

[swee'pea99]

Hands up if you say 'nah-STUR-tee-um'. Anybody? Ok, um, hands up if you've ever heard anybody say 'nah-STUR-tee-um'. You're British, right?

Strangely enough, the whole thing arose from something on Netflix last night, in which one of the characters referred to a 'nasturtium', and I turned to 'er indores and said, "nasturtium? Nasturtium?!" And she gave me that oh God here we go look but after consideration suggested maybe that's how Americans pronounce it. Which sounded at least vaguely plausible. So I just went to check and apparently it's not Americans who pronounce it like that: it's us. Who knew.

 

[Pblakeney]

View attachment 11556

Hands up if you say 'nah-STUR-tee-um'. Anybody? Ok, um, hands up if you've ever heard anybody say 'nah-STUR-tee-um'. You're British, right?

Strangely enough, the whole thing arose from something on Netflix last night, in which one of the characters referred to a 'nasturtium', and I turned to 'er indores and said, "nasturtium? Nasturtium?!" And she gave me that oh God here we go look but after consideration suggested maybe that's how Americans pronounce it. Which sounded at least vaguely plausible. So I just went to check and apparently it's not Americans who pronounce it like that: it's us. Who knew.

First time in my entire life that I have heard/read the word. In any context.

 

[briantrumpet]

View attachment 11556

Hands up if you say 'nah-STUR-tee-um'. Anybody? Ok, um, hands up if you've ever heard anybody say 'nah-STUR-tee-um'. You're British, right?

Strangely enough, the whole thing arose from something on Netflix last night, in which one of the characters referred to a 'nasturtium', and I turned to 'er indores and said, "nasturtium? Nasturtium?!" And she gave me that oh God here we go look but after consideration suggested maybe that's how Americans pronounce it. Which sounded at least vaguely plausible. So I just went to check and apparently it's not Americans who pronounce it like that: it's us. Who knew.

Proper fail there. The only difference is whether the first syllable is a schwa or not, and whether the R is sounded. They should have just checked the OED.

 

[briantrumpet]

 

[briantrumpet]

Must admit I've not tried Gemini.

 

[Rusty Nails]

Why don’t you ask the AI you use which one is the best. You’re bound to get an accurate, unbiased answer.

 

[PurplePenguin]

I asked Copilot. It didn't include itself in the answer. I use Copilot because it doesn't need an account.

📊 How to Decide Which Is “Best”​

• For coding: Claude Sonnet 4.5 leads benchmarks.
• For reasoning: Grok 4 shines with huge context windows.
• For multimodal tasks: Gemini 2.5 is strongest.
• For general versatility: GPT‑5 remains the most balanced.
• For open-source flexibility: LLaMA 4 is the go-to.

 

[icowden]

Must admit I've not tried Gemini.

i've tried it with coding. It can do a good job and it's able to absorb more date more easily as well as use external tools like canvas to assist with things like UI creation. But it's not as technically accurate as copilot. If often found Gemini can get stuck in a loop, unable to fix its code, whereas Copilot is much better at paring backand restructuring.

 

[Rusty Nails]

i've tried it with coding. It can do a good job and it's able to absorb more date more easily as well as use external tools like canvas to assist with things like UI creation. But it's not as technically accurate as copilot. If often found Gemini can get stuck in a loop, unable to fix its code, whereas Copilot is much better at paring backand restructuring.

All this stuff on AI help for coding makes me feel even more of a dinosaur than I already did. I must confess to being a computer programmer in 1968 and working in IT for seven years until I managed to make my escape. Ah, the good old days of basic assembler and decks of punched cards and going into work at night to get mainframe time to test your programs, plus double-checking your coding to make sure you had minimised the space taken by the program.
On the plus side those offices of ‘punch girls’ (sorry for the 70s casual sexism) were a pleasure to work with.

 

[icowden]

All this stuff on AI help for coding makes me feel even more of a dinosaur than I already did. I must confess to being a computer programmer in 1968 and working in IT for seven years until I managed to make my escape. Ah, the good old days of basic assembler and decks of punched cards and going into work at night to get mainframe time to test your programs, plus double-checking your coding to make sure you had minimised the space taken by the program.

My dad had used punch card systems, presumably part of his work on pulsed Nuclear Magnetic Resonance (NMR) relaxation techniques to study molecular motions in polymers for his PHD research. I think he also learned COBOL and FORTRAN.

 

[briantrumpet]

My dad had used punch card systems, presumably part of his work on pulsed Nuclear Magnetic Resonance (NMR) relaxation techniques to study molecular motions in polymers for his PHD research. I think he also learned COBOL and FORTRAN.

Fortran was what the clever people did at school where I was in the 70s: they wrote out all the code by hand, which was sent over to Bristol Poly to be punchcarded and run, the output came back for debugging, etc...