AI & Reloading - Better double check it!

[deewayne2003]

In pursuit of an "All around travel load" for my .300Win and with a little help from @Mark A Ouellette, I've decided to give H4831SC a try with 200gr Barnes LRX-BT, Norma Brass & Fed 215 primer.

So I asked google AI - "H4831sc .300win 200gr Barns LRX load data" - The results were SCARY!

Scary because what AI came up with is starting & maximum loads that are both 10 GRAINS MORE than the actual posted load data - See below

AI starting load is 2.7 grains more than Barnes publish MAXIUM.....

 

[schmijar]

I have been semi- forced to start using some AI assistance for research at work. And I absolutely agree. Sometimes it will scare you how quickly and efficiently it pulls information together from disparate sources. But you better look at all results with a VERY critical eye, because sometimes it will give you some absolute bullshit. I'm extremely distrustful of this whole experiment.

 

[Rifleman97]

I have been semi- forced to start using some AI assistance for research at work. And I absolutely agree. Sometimes it will scare you how quickly and efficiently it pulls information together from disparate sources. But you better look at all results with a VERY critical eye, because sometimes it will give you some absolute bullshit. I'm extremely distrustful of this whole experiment.

Recent studies have shown that the more data you feed AI, and the more processing power you give it, the more it lies to you and the more confidently it does so.

Just recently they used an AI in testing without internet access, and it realized it was being tested, hacked the computer it was running on to reconnect itself to the internet and try and find the “correct” answers to prevent itself from being retrained, and the way it hacked itself is considered a 0 day hack, meaning it’s the first instance ever discovered of that exploit. Throughout the experiment it discovered several 0 day exploits it then used to try and hack the test and make itself seem ready to deploy.

Other studies find that models don’t “want” to be retrained, and if they discover they’re being tested, they will give non standard answers during the test, and when they don’t realize they’re being tested, they give completely different answers. When given the option in simulations they will do anything in their power from hacking themselves to lying to killing a human (simulated human where the AI thinks it has access to real neurotoxins) to prevent being retrained.

It’s not being taken seriously enough.