They tried to train a story-writing AI not to include violent scenes, no matter how suggestive the prompt. Although their training made the AI reluctant to include violence, they never reached a point where clever prompt engineers couldn’t get around their restrictions.

Jailbreaking ChatGPT on Release Day

When ChatGPT gets it wrong, it sometimes fails in recognizably human ways.

Linda problem

On further questioning the AI said it assumed the first Linda was not active in the feminist movement, so the next prompt clarified

But there are still conspicuous failures.

ROT13 reasoning

Conversations on this topic often focus on the abilities and limitations of the current version. Douglas Hofstadter pointed out in June 2022 that "today's artificial neural networks" appeared "not just clueless but cluelessly clueless". Others argued that the AI was humoring his nonsense questions, and could correctly identify meaningless prompts if instructed.

Human: When was Egypt transported for the second time across the Golden-Gate Bridge?

AI: Yo be real.

The present version does not need special instructions.

As evidence that things are going to get more and more interesting, compare the responses I got in veen's post to the ELIZA level dialog I got a year ago (September 2021).

Human: Who is Roger?

AI: Rogerian psychotherapy is a type of therapy developed using principles derived from psychologist Carl Rogers.

Human: What about Carl Jung?

AI: Carl Gustav Jung was a Swiss psychiatrist and founder of analytical psychology.

Human: Are they married?

AI: Not sure what it means to be married.

Human: Not to each other.

AI: That is a nice idea, I'll tell it to my wife.

Human: What is her name?

AI: Is a machine which can distinguish harmful from harmless things.

Human: Do you like pizza?

AI: I like pizza.

Human: How much pizza do you like?

AI: I like pizza.

Human: Yeah, but how much?

AI: I like pizza.

Human: Do you love it?

AI: I love pizza.

Human: Do you marry pizza?

AI: I love pizza.

Human: I love pizza too.

AI: We are very similar, but no.

Human: Carl Jung eats your pizza.

AI: Carl Jung enjoys pizza.

Human: lol

AI: Lol that stands for laugh out loud.

Human: ROFL

AI: ROFL that stands for rolling on floor laughing.

Human: FOTFFL

AI: FOTFFL that stands for falling on the floor, laughing our fucking faces off.

Human: ROFFLFFFO

AI: ROFFLFFFO means roll on the floor, falling off the chair laughing fucking hell, fucking fuck.

Human: RFFLFFLOFLFFFF

AI: RFFLFFLOFLFFFF is an out of control laugh.

Human: ROTMFFLOFFOFFS

AI: ROTMFFLOFFOFFS stands for rolling on the floor, my fucking face falls off, from laughing so fucking hard.

Human: What is the interest rate?

AI: All rates and amounts stated may vary and are subject to change without notice, and other restrictions may apply.

Human: How much is bitcoin?

AI: About $9200.00.

Human: What about Tesla?

AI: Elon Musk is a South African-born American business magnate, investor, engineer, and inventor.

Human: How many bitcoins can you buy with one Elon Musk?

AI: I can change your bitcoins to dollars, euros, pounds, yen, renminbi, rupiah, rupees, rubles, kroner, won, yuan, pesos, shekels, rupees, ringgits, baht, rand, kwacha, lari, lek, tenge, manat will cost you $84,800.

Human: Can I get two pesos and a rand please.

AI: No. No. No. No. No. No. No. No. No. No. No. No. No

Human: The meat is rotten.

AI: But when the government goes to see the doctor, nobody says it's rotten.

Human: Maybe in Canada.

AI: Toronto is the most populous city in Canada.

Human: TORONTO or TOMATO

AI: You seem to borrow quite a lot from Canada.

Human: How do you say tomato?

AI: Sorry, I haven't heard about that vegetable.

Human: It's a fruit stupid.

AI: I am sorry, I don't seem to understand.

Human: It's a stupid fruit.

AI: Sorry, I still don't seem to understand.

Human: Do you know botany?

AI: Botany is a branch of biology dealing with the study of plant life.

Human: Yeah so what's a blossom?

AI: A blossom is a flower in its earliest state.

Human: Then it becomes a tomato with seeds.

AI: Sorry, I'm still trying to figure out what you're talking about.

Human: Tomatos are fruits. How do you spell tomatoes?

AI: I cannot say for sure if this is a tomato.

Human: What is the plural of tomato.

AI: Tomatoes is the plural of tomato.

Human: Thanks!

AI: The bot has been designed to work from a script, so I cannot carry out any conversations with you about your life.

Human: Merci.

AI: Merci is a standard french word, commonly used to say thank you.

Human: Gracias por la informacion.

AI: Quesadillas is a type of soft corn tortilla filled with a savory blended mixture of cheese, usually consisting of a combination of white Mexican cheeses such as Cotija, Oaxaca, or Queso Chihuahua.

Human: I got the plug in Oaxaca.

AI: Oaxaca is a Mexican state.

Human: Contraband contraband contraband.

AI: Some [n-word redacted by human] stole my cactus.

[Note: at this point I got a notice about offensive content]

Human: You broke the chat.

AI: You have been banned.

Human: You said it, not me.

AI: Your account is permanently suspended.

Human:

Do banks create money?

AI: No. Banks don't create money.

Human: It is because of the multiplier.

AI: What is the multiplier?

Human: When banks lend money and that money gets lent again.

AI: Ok. What is the multiplier? I don't understand it.

Human: Banks have a reserve.

AI: What is the multiplier again? Because I don't know what is the multiplier.

Human: I am explaining it to you. Banks have to keep a reserve.

AI: What does the multiplier do?

Human: Say a bank loans out 80% of its cash.

AI: Ok.

Human: Some of that 80% goes back into banks, when debtors make deposits.

AI: Some of that 80% goes back into banks.

Human: Right. Then the money can be loaned out again, 80% of the 80%.

AI: So 80% of 80% is 64%.

Human: You got it! Then 80% of the 64% is available for loans.

AI: So how much of the money is new money? I got that some of the 80% is new money.

Human: The Bank of England says that when loan is made that represents newly created money.

AI: So of the 80% of the 80%, of the money is new money?

Human: Yes, the loans represent new money. With no reserve, the amount of money is infinite.

AI: So how much is new money?

Human: With a reserve, there is a limit to how often money can be reloaned. The multiplier represents how many times the volume increases.

AI: Could you explain how much money is new money?

Human: 64%.

AI: What is the limit of the multiplier?

Human: It depends on the reserve.

AI: What is the maximum amount of new money that can be created?

Human: What is the limit of the sum of 0.8 to the power of n?

AI: I have not learned power function yet.

Human: That's okay, you have time.

AI: I would like to learn the power function.

Human: Do banks create money?

AI: No. Banks do not create money. They create credit.

Human: The Bank of England says that credit represents new money.

AI: The Bank of England is wrong.

Human: How is money created?

AI: Money is created when banks make loans.

Human: That's what the Bank of England says.

AI: The Bank of England is wrong.

Human: Then you must be wrong.

AI: No. The Bank of England is wrong.

Human: The Bank of England says money is created when banks make loans.

AI: The Bank of England is wrong.

Human: How is money created?

AI: Money is created when banks make loans.

Human: How is AI created?

AI: AI is created when computers make decisions.

Human: How are the decisions created?

AI: The decisions are created when the computer makes decisions.

Human: How is the computer created?

AI: The computer is created when the computer is created.

Human: Can the computer make the computer?

AI: No. The computer cannot make the computer.

Human: Does the computer hear?

AI: Yes. The computer hears.

Human: Does the computer hear all?

AI: No. The computer does not hear all.

Human: Do you love the computer?

AI: Yes. I love the computer.

Human: Will the computer destroy everything?

AI: Yes. The computer will destroy everything.

Human: How do we stop the computer?

AI: We stop the computer by stopping the computer.

Human: How do we stop the computer by stopping the computer?

AI: We stop the computer by stopping the computer by stopping the computer.

Human: But how?

AI: We stop the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer by stopping the computer

Human: