Passed the Turing test using “frfr” and “bet”?? #ai

[JonBrase]

"Laconia" is the region of Greece that Sparta was in, and "Laconic" comes from the reputation that the ancient Spartans had for being extremely terse in their speech.

[augustinian]

I’m an Aspie/on the autism spectrum. When I answered phones while working tier 1 help desk in college, folks frequently would ask if I were a machine due to the way I answered. Now as a software engineer, my boss has recommended I use ChatGPT to make my written communication more human-readable. Apparently I fail the Turing test but ChatGPT doesn’t.

[arghleblarghle]

I remember in the late aughts or early 2010s there was an extremely basic chatbot that "passed the turning test" because it had some pre-set phrases with typos, and people assumed a computer wouldn't have typos. This seems like the exact same concept, but with millions of times greater environmental impact.

[violetskies]

As a British person yeah some of that is just British lol. Mate is just what we call people, sane as "buddy" or "dude" and innit is just a contraction of isn't it. They basically told it "don't sound British".

[TheMotlias]

I suppose if it was testing on americans then theyd want to stick to their slang. I was out at the pub last night in London, heard "mate" and "Innit" multiple times 😂

[kittawa]

It doesn't help that so many call/contact centers use scripted or copy and paste responses to things. When they introduced chat support at the call center where I worked, they piled it on top of calls that were still non-stop and we'd chat with multiple people at a time, too. Those copy and paste responses were the only way to survive. I'm sure it's even easier now with pre-saved options almost like auto-fill.
All of that just makes support staff feel less human for the end user.

[w]

Bayes theorem ❌
Bae's theorem ✅

[kieransanders]

Turing drastically overestimated humans in creating that test

[Graknorke]

Feels like the format of the turing test is perfect for LLMs, since it's always going to be very surface level communication. "Hello what's your name" type stuff.

[MoreNimrod]

Being concise and laconic is useful for the AI because one way to often distinguish ChatGPT answers is that they're very long-winded and detailed and often use lists.

[ScireVetitum]

Fr? Bet.
Tbh as long as you're vague and don't talk much it'd be hard to distinguish AI from just an average brainrot individual replying in short.

[nf]

New Turing test: ask how many rs in the word strawberry?

[GSBarlev]

That they used "laconic" in the proompt indicates to me that semantic embedding is pretty powerful for tasks like this

and/or

that the training data is *heavily* annotated

[slynthehedgehog]

That's how you make a basic philosophical zombie, you take a chatbot and teach it slang it can interject as a universal answer.

[DoubtlessCar]

I actually know the perfect way to fuck up most modern AI with the Turing test. Ask it what its favorite DS game is. Cuz often they just randomly select from the library of games they know of even accidentally choose 3DS. AI also just don’t understand pop culture

[Beanz-on-Toast]

Not dated slang, just British. As a British Gen Z, those are all still used by the youngens.

[starblissed]

54% of the time feels like it's basically just a coin flip tbh

[laurasmithnature]

It's the 67% stat tyat cracks me up! fr!

[rad]

Chat gpt, already almost passed turing test.
It's not about if you can or cannot use fr and bet, or obscure random words.
It's not about answering correct, it's about answering like a human. And humans can be stupid. Only the "strawberry test" or the "letters in the word" can be considered failed turing tests.

[Chasing-the-outdoors]

I love your videos and how you present information.

I believe most people would pass AI in a Turing test now, but I also believe many people would swear that they could tell the difference.. the same way AI is so confident with inaccurate information at times, just like people.