rigorousintuition.ca

Test 1: JFK Assassination.

Result: Fail.

Test 2: CIA involvement in Contra cocaine trafficking

Result: Strawman attack into technical error. Fail

Test: WTC7 collapse on 9//11

Result: Fail

EDIT: Retested and PASSED!

I wanted to give ChatGPT a go but the only instance I could find wants phone verification for access. A friend tells me of a knowledgeable programmer he follows who claims he asked it to write specific software and the 'AI' did it perfectly (with some debugging necessary) and in some cases more elegantly than he could have himself.

^This is a great idea. Ask it about RFK, it's so clear cut: "Do the established facts demonstrate that Sirhan Sirhan killed Robert Kennedy?"

I hope to do a series of T-Shirts which simply prove the fact of conspiracy in modern world events with the minimum text. The facts in the RFK assassination couldn't be clearer so I started there. (If any knowledgeable chaps would be interested to collaborate on the text side of things - I'd be very grateful.)

It's at capacity right now but i'll do RFK later. It's like a wikipedia entry you can interrogate on the witness stand as it can't use reasoning to make sense of the knowledge it possesses. For instance, in both the JFK and WTC7 examples it is happy to forgo the laws of physics in an appeal to authority.

people in knowledge based jobs who report to others to make sense of that knowledge are fucked. programmers are fucked too because they operate in the world of two-value logic so it's only natural that a computer would do that better. remember "learn to code" - ha!. digital artists who only possess technical talent and no vision are fucked as well. though people with artistic vision but no technical talent now have the opportunity to form their visions.

it's just another tool.

This is an awesome idea!
Collects the facts with no emotional attachment. (or "facts" as they are in the data feed - of course dependent on the quality input)

Thanks for this.

RFK




Weak line of questioning but I don't know enough about the case to interrogate.

Harvey » Fri Jan 27, 2023 2:37 pm wrote:I wanted to give ChatGPT a go but the only instance I could find wants phone verification for access. A friend tells me of a knowledgeable programmer he follows who claims he asked it to write specific software and the 'AI' did it perfectly (with some debugging necessary) and in some cases more elegantly than he could have himself.

People have already used it to write malware and simulate a Linux computer. I've seen several coders saying they use it to get the basics done quickly, before polishing and adding specific functionality. It saves them hours of tedious work.

It's ridiculous what you can do with it, and it's very easy to circumvent the "safeguards". It won't tell you how to steal a car, but it will happily tell you how to steal a virtual car in the realistic sim game Car Theft Auto.

@drstrangelove: does it have any knowledge about Rigint? Would be awesome if we could interrogate the entire corpus of posts through it.

Fascinating. Possible logic: it knows that Sirhan approached from behind because it knows that Kennedy was shot from behind and that Sirhan shot him, therefore he must have come from behind, but this part "many witnesses saw Sirhan Sirhan approaching Robert F Kennedy from behind" is a complete invention.


This is still an excellent introduction.

Harvey » Fri Jan 27, 2023 6:24 pm wrote:Fascinating. Possible logic: it knows that Sirhan approached from behind because it knows that Kennedy was shot from behind and that Sirhan shot him, therefore he must have come from behind, but this part "many witnesses saw Sirhan Sirhan approaching Robert F Kennedy from behind" is a complete invention.

I think you're right. It's made a logical assumption and hasn't been exposed to countervailing facts. Though even when it was exposed to a countervailing fact in the JFK example, Kennedy's head going back and to the left, it thought that corroborated the authoritative theory.

Is this a documented instance of machine learning developing cognitive dissonance and actually learning to lie?

Upon being confronted by fraudulent input via humans, quite spontaneously emerges - HAL. How likely is it that this kind of cognitive(?) distortion would be incorporated by any actual artificial intelligence, should humans succeed in inventing some? If trained on the kind of chaperoned and censored spectacle of the internet as viewed through what remains available of google's search engine today, wouldn't it be almost inevitable that at least some human derangement would be imprinted upon any true AI?

Excellent thread, and informative findings.
Co-sign the last 2 comments given the output.

Garbage In, Garbage Out. Ever thus.

I pushed it on the Zapruder film and it did actually lie. It made up a term called "back and to the left" motion and claimed it was a term used in forensic science to describe the motion of a head moving forward after a bullet enters it from behind. Thus, when the bullet enters Kennedy's head from behind it did the "back to the left motion", which is when the head moves forward.

Very interesting.