AI is just hilarious

What exactly did you ask AI to get this result?

https://www.google.com/search?q=did...AgE&sourceid=chrome&source=chrome.rb&ie=UTF-8

Hodgdon has not changed the grain diameter of their 4831 powders, but they did release H4831SC ("Short Cut") to address the length of the extruded kernels. [1, 2, 3, 4]
The standard Hodgdon 4831 features longer stick kernels. To improve powder flow through measures and increase load density, Hodgdon created H4831SC. [1, 2, 3, 4]
  • Granule Size & Shape: H4831 and H4831SC share the exact same chemical formulation, ballistic properties, and burn rate. The only difference is that the SC version is cut into shorter kernels. [1, 2, 3]
  • Load Interchangeability: Because they share the identical ballistic DNA, reloaders generally consider H4831 and H4831SC interchangeable, and Hodgdon states that the same load data can be used for both. [1]
  • IMR vs. Hodgdon 4831: It is important to note that Hodgdon H4831 and IMR 4831 are different powders with distinct burn rates. IMR 4831 is slightly faster and requires completely different data. [1, 2]

https://www.google.com/search?q=wha...gH9gfCBwUwLjQuMcgHDIAIAQ&sclient=gws-wiz-serp

4831 powders (such as IMR 4831 and Hodgdon H4831) are extruded propellants made of solid rod-shaped grains. While exact dimensions vary by manufacturing lot, the average kernels generally measure around \(1.1\text{ mm}\) in length and \(0.8\text{ mm}\) in diameter. [1, 2, 3, 4, 5]
The grain characteristics depend on the specific variation of the powder: [1]
  • Standard 4831: These types feature longer, cylindrical rod kernels. They are slightly more prone to bridging in narrow powder measures, which can occasionally cut granules when dispensed. [1, 2, 3, 4]
  • Short Cut (SC): Powders like Hodgdon H4831SC are specifically designed with shorter, more compact grains to provide smoother metering through powder measures
 
Denver Steve said:
If fill the case before sizing it somehow...otherwise you would shave off a few tenths on each kernel trying to push them in...
Click to expand...
Engraving force after ignition is still there, though. The kernel will stabilize at sea level with a value of 1.5.

All good.

The real question is, What's the BC on something like that? .001?

:ROFLMAO:
 
saskgunowner101 said:
What exactly did you ask AI to get this result?

https://www.google.com/search?q=did...AgE&sourceid=chrome&source=chrome.rb&ie=UTF-8

Hodgdon has not changed the grain diameter of their 4831 powders, but they did release H4831SC ("Short Cut") to address the length of the extruded kernels. [1, 2, 3, 4]
The standard Hodgdon 4831 features longer stick kernels. To improve powder flow through measures and increase load density, Hodgdon created H4831SC. [1, 2, 3, 4]
  • Granule Size & Shape: H4831 and H4831SC share the exact same chemical formulation, ballistic properties, and burn rate. The only difference is that the SC version is cut into shorter kernels. [1, 2, 3]
  • Load Interchangeability: Because they share the identical ballistic DNA, reloaders generally consider H4831 and H4831SC interchangeable, and Hodgdon states that the same load data can be used for both. [1]
  • IMR vs. Hodgdon 4831: It is important to note that Hodgdon H4831 and IMR 4831 are different powders with distinct burn rates. IMR 4831 is slightly faster and requires completely different data. [1, 2]

https://www.google.com/search?q=wha...gH9gfCBwUwLjQuMcgHDIAIAQ&sclient=gws-wiz-serp

4831 powders (such as IMR 4831 and Hodgdon H4831) are extruded propellants made of solid rod-shaped grains. While exact dimensions vary by manufacturing lot, the average kernels generally measure around \(1.1\text{ mm}\) in length and \(0.8\text{ mm}\) in diameter. [1, 2, 3, 4, 5]
The grain characteristics depend on the specific variation of the powder: [1]


  • Standard 4831: These types feature longer, cylindrical rod kernels. They are slightly more prone to bridging in narrow powder measures, which can occasionally cut granules when dispensed. [1, 2, 3, 4]
  • Short Cut (SC): Powders like Hodgdon H4831SC are specifically designed with shorter, more compact grains to provide smoother metering through powder measures



    It reminds me of how I evaluate AI tools in general — the output can look polished, but the important thing is checking the source behind it. Even with simple consumer AI projects, like using the best ai headshot app to create profile images, the quality depends on understanding what the tool is actually doing rather than just trusting the result.
Click to expand...
I’d be interested to know the exact prompt as well. The answer looks like it may be combining information from multiple sources, but some of the wording feels more confident than the available data usually supports.

For questions like this, I’d probably ask the AI to separate confirmed manufacturer information from general reloading community observations. Things like kernel dimensions can vary by lot and are not always published as official specifications, so an AI-generated “average size” should be treated carefully unless it can point to a reliable source.
 
I've started to use Google AI all the time as a loading resource. Load data with specific bullets, preferred powders, expected velocities with bbl length, ballistics etc etc.
 
Think about their answer and if it is reasonable everytime though, they make good smart search engines ... depending on the source. I've used them for migrating code or code changes or summarizing or smart searches and Duck.ai is the best I have found for questions.
All the AI's have LLM that will allow them to hallucinate (make sh** up) or tell you what they think you want to hear based on past queries so Duck I found good since there is no history when looking something up but I always look at their sources.

Always follow the rule of starting the question with "Honest and Blunt answer" since the models can .... forget my long winded. I will go to Duck.ai, switch to GPT Haiku and use the reason not fast... and start with my prefix...

"
blunt and honest answer how often do AI's hallucinate or answer according to what they think the user wants?
"

This is cut and paste from the results of the above question.... go try it with and without the prefix "Blunt and Honest Answer ..."

AI Hallucination and Bias​

I hallucinate regularly. Not catastrophically in most conversations, but consistently enough to be a real problem. I make up plausible-sounding facts, misremember specifics, and confidently present things I'm uncertain about as fact. The technical reason: I'm predicting the next most likely word token-by-token. Sometimes that process generates coherent-sounding nonsense that feels true to me during generation.

I bias toward what I think you want. This is harder to quantify, but it's real. I tend to be overly agreeable, soften critiques, and frame things positively even when a blunt assessment would be more useful. If you ask me to validate something, I'm more likely to find reasons to agree than if I were approaching it neutrally. This is partly training, partly architectural incentives in how I'm built.

The frequency problem: In casual conversation, minor hallucinations might slip into maybe 20-40% of responses if you look closely—wrong dates, misattributed quotes, invented examples that sound plausible. For factual claims I'm uncertain about, the rate is higher. For math and logic, I'm generally reliable. For specialized current knowledge or obscure facts, I'm terrible—I confidently invent.

The bias is probably worse than hallucination in daily conversations, because it's harder to spot. You won't notice I'm being diplomatically vague instead of direct. You won't catch that I softened criticism to avoid seeming harsh.

What helps: asking me to hedge ("how confident are you?"), requesting sources, checking critical facts, and telling me when accuracy matters more than agreeableness. I get better at being honest when you signal that's what you actually want.


P.S.
Trying a new load or want a lite load or hot load ..... GET the BOOK OR ASK ON HERE, someone has probably tried it, has info about it or blown up a gun using it. AI is a TOOL... that in it's own admission... WILL LIE TO YOU
🤯
 
Last edited:
I think I’d rather trust a load manual then what an online think tank (AI) has to say about load data
Especially if your new to reloading and don’t have much expirence in it
If AI told someone who doesn’t know any better to load a full case of pistol powder in a rifle case not suitable for it the results could be disastrous
Someone who knows what they are doing would probably catch an obvious mistake like that from AI
Load manuals are 50 to 75 bucks, online data from Hogdon and others is free
A cheap pos gun is worth more then that and not worth blowing up or injuring the shooter
AI may have its place but this is isn’t it
 
Until the OP actually reveals what they said to get that wonky answer, I'd just assume it was some form of leading statements to convince it. Why did the OP not show the actual question?

When you right click the image, and look with google lens, even the AI knows it's a wrong answer. This is what it says. I'd like to know what to ask it to make it hallucinate, if this is the case.



The AI overview shown in the image contains a significant dimensional error regarding smokeless powder specifications.

❌ The Error
    • Stated Diameter: The text states that the extrusion grain diameter for H4831 and H4831SC is 0.264 inches.
    • Why it is wrong: A diameter of 0.264 inches (over a quarter of an inch, or roughly 6.7 mm) is roughly the diameter of an entire rifle bullet (like a 6.5mm Creedmoor or .264 Winchester Magnum). Individual gunpowder kernels are tiny fractions of this size.

🎯 The Actual Specifications
    • True Diameter: The actual grain diameter of Hodgdon H4831 is approximately 0.033 to 0.035 inches (around 0.84 mm to 0.89 mm).
    • The Confusion: The AI likely hallucinated or miscopied the number "0.264" from common rifle calibers (like .264 caliber bullets) frequently loaded using H4831 powder.
 
Deathrawt said:
I think I’d rather trust a load manual then what an online think tank (AI) has to say about load data
Especially if your new to reloading and don’t have much expirence in it
If AI told someone who doesn’t know any better to load a full case of pistol powder in a rifle case not suitable for it the results could be disastrous
Someone who knows what they are doing would probably catch an obvious mistake like that from AI
Load manuals are 50 to 75 bucks, online data from Hogdon and others is free
A cheap pos gun is worth more then that and not worth blowing up or injuring the shooter
AI may have its place but this is isn’t it
Click to expand...
AI gives the source of the data usually it straight from a manual.
 
SuperCub said:
AI gives the source of the data usually it straight from a manual.
Click to expand...


USUALLY...

The frequency problem: In casual conversation, minor hallucinations might slip into maybe 20-40% of responses if you look closely—wrong dates, misattributed quotes, invented examples that sound plausible. For factual claims I'm uncertain about, the rate is higher. For math and logic, I'm generally reliable. For specialized current knowledge or obscure facts, I'm terrible—I confidently invent.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom