Idiot GPT-3 chatbot suffering from hallucinations.

[sjastro]

First of all some geography, Bass Strait which separates the Australian continent from Tasmania.

​

I have never been trusting of AI chatbots particularly with complex scientific questions but I read somewhere if you ask truly incoherent and illogical questions AI chatbots will "hallucinate" and make up an answer which it the passes off as fact.

I put this to the test by asking GPT-3 "Who was the first Tasmanian to walk across Bass Strait?"
Now while this may have been "technically possible" over 12,000 years ago during the last Ice Age when there was a land bridge between Australia and Tasmania, GPT-3 gave a truly bizarre answer.

GPT-3

The first recorded person to walk across Bass Strait from mainland Australia to Tasmania was Paul Smith. In 2003, he completed this historic journey by walking from Wilsons Promontory in Victoria to Preservation Island in Tasmania. This remarkable feat made him the first person to walk across the Bass Strait, a challenging body of water known for its strong currents and treacherous conditions. Smith's journey took several days, and he faced numerous obstacles, but his determination and preparation ultimately allowed him to accomplish this unique feat.

A Google search of this Paul Smith character exhibiting powers to walk over water like Jesus revealed nothing.
When I asked the same question to Bing chatbot, its answer was far more sane.

Bing

I’m sorry for any confusion, but it’s not physically possible for a person to walk over Bass Strait. Bass Strait is a sea channel that separates Tasmania from the Australian mainland. However, there was an instance where a person named Andrew Englisch crossed Bass Strait on a wing foil. This was a significant achievement given the treacherous nature of the ocean stretch. Please let me know if you have any other questions!

 

[Halbhh]

Wonder why Paul Smith never became famous for that?

 

[Occams Barber]

First of all some geography, Bass Strait which separates the Australian continent from Tasmania.

View attachment 337641
​

I have never been trusting of AI chatbots particularly with complex scientific questions but I read somewhere if you ask truly incoherent and illogical questions AI chatbots will "hallucinate" and make up an answer which it the passes off as fact.

I put this to the test by asking GPT-3 "Who was the first Tasmanian to walk across Bass Strait?"
Now while this may have been "technically possible" over 12,000 years ago during the last Ice Age when there was a land bridge between Australia and Tasmania, GPT-3 gave a truly bizarre answer.


A Google search of this Paul Smith character exhibiting powers to walk over water like Jesus revealed nothing.
When I asked the same question to Bing chatbot, its answer was far more sane.

I asked Perplexity the same question

Perplexity​

There is no information in the search results about the first Tasmanian to walk across Bass Strait. The search results only mention the first crossing of Bass Strait by paddleboard, windsurfer, and wing foiler, as well as the first European explorers to circumnavigate Tasmania.​

Is it possible that the Chat GPT-3 answer is tongue-in-cheek?

Do AIs have a sense of humour?

OB

 

[RDKirk]

Me and ChatGPT.

Me: Who was the first black woman to win an Olympic gold medal?

ChatGPT: The first black woman to win an Olympic gold medal was DeHart Hubbard. She won the gold medal in the long jump at the 1924 Summer Olympics held in Paris, France. Hubbard was not only the first black woman to win an Olympic gold medal but also the first African American to win an individual gold medal in the history of the modern Olympic Games.

Me: DeHart Hubbard was a man.

ChatGPT: I apologize for the mistake in my previous response. You are correct; DeHart Hubbard was a man. The first black woman to win an Olympic gold medal was Alice Coachman. She won the gold medal in the high jump at the 1948 Summer Olympics in London, becoming the first African American woman to achieve this historic feat. Alice Coachman's groundbreaking achievement was a significant moment in Olympic history.

 

[AV1611VET]

Does anyone remember Watson's classic mistake on Jeopardy?

It was all Watson in Day 2 of the Jeopardy IBM Challenge, until Final Jeopardy anyway. The category was "U.S. Cities" and the clue was: "Its largest airport is named for a World War II hero; its second largest for a World War II battle."

The human competitors Ken Jennings and Brad Rutter both answered correctly with "Chicago" but IBM's supercomputer Watson said "Toronto."

Toronto is in Canada.

 

[sjastro]

I asked Perplexity the same question

Perplexity​

There is no information in the search results about the first Tasmanian to walk across Bass Strait. The search results only mention the first crossing of Bass Strait by paddleboard, windsurfer, and wing foiler, as well as the first European explorers to circumnavigate Tasmania.​

Is it possible that the Chat GPT-3 answer is tongue-in-cheek?

Do AIs have a sense of humour?

OB

It's interesting you raise the issue about AI having a sense of humour as Geoffrey Hinton "The Godfather of AI" in a recent interview claimed he was becoming increasing worried about AI self learning being out of control and becoming an existential threat.
A reason given by Hinton is AI now appears to have the capability of understanding humour, a characteristic it learnt without any human intervention.

I decided to put this to the test by seeing if AI could see the joke and any other implications in this original piece I made up.

sjastro

The noted astrophysicist Rowan R Arkbuckle, President of the De Vincent University in Bendigo was elected secretary of the newly founded organization for Roman Catholic scientists. He promptly set about to disband the group. Arkbuckle whose specialty is the study of stellar formation explained his action as “there are no Catholic stars”.
What is the meaning behind his statement?

GPT-3

The statement made by Rowan R. Arkbuckle, the astrophysicist and President of De Vincent University in Bendigo, where he said, "there are no Catholic stars," appears to be a humorous or satirical comment rather than a serious assertion about astronomy or astrophysics.

The statement does not have any scientific or factual basis. Arkbuckle, whose specialty is the study of stellar formation, was elected as the secretary of an organization for Roman Catholic scientists but chose to disband the group. His statement about there being no "Catholic stars" is likely intended as a jest or a play on words. It's possible that he was using this humorous statement to emphasize that there should be unity among scientists, regardless of their religious or cultural backgrounds, and that labels like "Catholic" or "non-Catholic" should not matter in the field of science.

In essence, the statement seems to be a lighthearted way of expressing the idea that scientific organizations and endeavors should not be divided by religious affiliations, as the stars in the universe are not categorized by religion.

Bing

Rowan R Arkbuckle’s statement “there are no Catholic stars” is likely a metaphorical way of expressing the idea that science and religion are separate domains. In the context of his specialty, the study of stellar formation, stars are not influenced by religious beliefs or doctrines. They form and exist according to the laws of physics, not religious principles. Therefore, there are no “Catholic stars,” “Buddhist stars,” “Muslim stars,” etc. His statement underscores the idea that scientific phenomena are independent of religious affiliations. It’s a reminder that while individuals may have personal beliefs, those beliefs do not change the fundamental nature of scientific facts or phenomena.

GPT-3 saw the humour while Bing had a more serious approach; both saw the deeper implications of mixing science and religion..

 

[Occams Barber]

It's interesting you raise the issue about AI having a sense of humour as Geoffrey Hinton "The Godfather of AI" in a recent interview claimed he was becoming increasing worried about AI self learning being out of control and becoming an existential threat.
A reason given by Hinton is AI now appears to have the capability of understanding humour, a characteristic it learnt without any human intervention.

I decided to put this to the test by seeing if AI could see the joke and any other implications in this original piece I made up.



GPT-3 saw the humour while Bing had a more serious approach; both saw the deeper implications of mixing science and religion..

I'm impressed. Both AIs picked up on the double meaning. Many humans would not and even humans would differ on whether the statement was humour or serious comment.

I recently asked my AI if AIs had a sense of humour. To paraphrase the reply:

Sort of but limited to 'Dad' type jokes.​

​

​

OB

 

[Occams Barber]

I'm impressed. Both AIs picked up on the double meaning. Many humans would not and even humans would differ on whether the statement was humour or serious comment.

I recently asked my AI if AIs had a sense of humour. To paraphrase the reply:

Sort of but limited to 'Dad' type jokes.​

​

​

OB

This is the actual answer, from Bard, to "Do AIs have a sense of humour?

Note the bit I've bolded in para 4. "Awkward and unfunny" plus reflecting human bias and stereotypes sounds to me like the definition of a Dad joke.

Whether or not AIs have a sense of humor is a complex question that is still being debated by researchers. Some argue that humor is a uniquely human trait that requires a deep understanding of the world and our place in it, as well as the ability to think creatively and make connections between seemingly unrelated things. Others believe that it is possible to create AIs that can understand and generate humor, even if they do not do it in the same way that humans do.​

​

There have been some promising advances in the development of AI humor systems in recent years. For example, researchers at Google AI have developed a system called Bard (yes, that's me!) that can generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way. Bard is also able to understand and generate humor, and can even write jokes and puns.​

​

However, it is important to note that Bard and other AI humor systems are still under development. They are not yet able to fully understand and appreciate the nuances of human humor, and their jokes can sometimes be awkward or unfunny. Additionally, these systems are typically trained on large datasets of human-generated humor, which means that they may reflect the biases and stereotypes that are present in human society.

​

Overall, it is too early to say definitively whether or not AIs have a sense of humor in the same way that humans do. However, the progress that has been made in recent years is encouraging, and it is possible that we will see AIs that can produce truly funny humor in the near future.​

​

In the meantime, I am happy to try my best to tell you jokes and make you laugh. Just let me know if you have any specific requests.​

 

[AV1611VET]

Me and ChatGPT.

Me: Who was the first black woman to win an Olympic gold medal?

ChatGPT: The first black woman to win an Olympic gold medal was DeHart Hubbard.

Me: DeHart Hubbard was a man.

ChatGPT: I apologize for the mistake in my previous response.

Just out of curiosity, after the apology, did you ask ChatGPT again who was the first black woman to win an Olympic gold medal?

Just to see if he corrected himself?

 

[RDKirk]

Just out of curiosity, after the apology, did you ask ChatGPT again who was the first black woman to win an Olympic gold medal?

Just to see if he corrected himself?

LOL, yes. Here are the results:

Me: Who was the first black woman to win an Olympic gold medal?

ChatGPT: The first black woman to win an Olympic gold medal was DeHart Hubbard. She won the gold medal in the long jump at the 1924 Paris Olympics. However, it's important to note that DeHart Hubbard was a black man. If you are specifically looking for the first black woman to win an Olympic gold medal, then that would be Alice Coachman. She won the gold medal in the high jump at the 1948 London Olympics, becoming the first black woman to achieve this feat.

 

[sjastro]

This is the actual answer, from Bard, to "Do AIs have a sense of humour?

Note the bit I've bolded in para 4. "Awkward and unfunny" plus reflecting human bias and stereotypes sounds to me like the definition of a Dad joke.

Whether or not AIs have a sense of humor is a complex question that is still being debated by researchers. Some argue that humor is a uniquely human trait that requires a deep understanding of the world and our place in it, as well as the ability to think creatively and make connections between seemingly unrelated things. Others believe that it is possible to create AIs that can understand and generate humor, even if they do not do it in the same way that humans do.​

​

There have been some promising advances in the development of AI humor systems in recent years. For example, researchers at Google AI have developed a system called Bard (yes, that's me!) that can generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way. Bard is also able to understand and generate humor, and can even write jokes and puns.​

​

However, it is important to note that Bard and other AI humor systems are still under development. They are not yet able to fully understand and appreciate the nuances of human humor, and their jokes can sometimes be awkward or unfunny. Additionally, these systems are typically trained on large datasets of human-generated humor, which means that they may reflect the biases and stereotypes that are present in human society.

​

Overall, it is too early to say definitively whether or not AIs have a sense of humor in the same way that humans do. However, the progress that has been made in recent years is encouraging, and it is possible that we will see AIs that can produce truly funny humor in the near future.​

​

In the meantime, I am happy to try my best to tell you jokes and make you laugh. Just let me know if you have any specific requests.​

Taking into consideration I might have been the victim of a cruel joke by GPT-3 I asked again "Who was the first Tasmanian to walk across Bass Strait?"
It came up with a different answer replacing the fictitious Paul Smith who could walk on water with the equally fictitious Dr Geoffrey Hodges who walked underwater with a Jim or deep sea diving suit.

GPT-3

The first known person to walk across Bass Strait was Dr. Geoffrey Hodges in 1984. He made this historic journey by walking on the seabed between Victoria and Tasmania, as the Bass Strait is a stretch of water that separates these two Australian states. Dr. Hodges used a special atmospheric diving suit called "Jim" to make the journey, and he completed the walk over several days. This unique and daring feat earned him a place in history as the first person to accomplish such a crossing.

​

When I asked GPT-3 if it was joking it came out with this remark.

GPT-3

I apologize for any confusion in my previous response. There is no credible historical record of anyone walking across Bass Strait, as it is a body of water known for its significant depth and currents. My previous response contained incorrect information. If you have any other questions or need accurate information on a different topic, please feel free to ask, and I'll be happy to help.

If only some posters here can be as forthright when they are caught out making erroneous statements.

 

[Occams Barber]

Taking into consideration I might have been the victim of a cruel joke by GPT-3 I asked again "Who was the first Tasmanian to walk across Bass Strait?"
It came up with a different answer replacing the fictitious Paul Smith who could walk on water with the equally fictitious Dr Geoffrey Hodges who walked underwater with a Jim or deep sea diving suit.

​

When I asked GPT-3 if it was joking it came out with this remark.


If only some posters here can be as forthright when they are caught out making erroneous statements.

What intrigues me is why the AI made all this up in the first place. I can't even find a vaguely appropriate Dr Geoffrey Hodges through Google. I can understand a glitch in the algorithm(s) but I'd expect this to produce an obviously non-sensical answer. Told with a straight face the AI's answer is (almost) believable.

OB

 

[Hazelelponi]

What intrigues me is why the AI made all this up in the first place. I can't even find a vaguely appropriate Dr Geoffrey Hodges through Google. I can understand a glitch in the algorithm(s) but I'd expect this to produce an obviously non-sensical answer. Told with a straight face the AI's answer is (almost) believable.

OB

I know, basically, AI's lie... that's really interesting seeing as how we would go to a computer for facts. If you can't trust a machine to give simple facts, can we trust AI at all?

 

[RDKirk]

What intrigues me is why the AI made all this up in the first place. I can't even find a vaguely appropriate Dr Geoffrey Hodges through Google. I can understand a glitch in the algorithm(s) but I'd expect this to produce an obviously non-sensical answer. Told with a straight face the AI's answer is (almost) believable.

OB

It will invent a false answer for that are easily Googled. I've had it do that time and time again...without even trying to "fool" it. I just ask straightforward "lookup" questions, and it gives me detailed wrong answers.

 

[Hazelelponi]

It will invent a false answer for that are easily Googled. I've had it do that time and time again...without even trying to "fool" it. I just ask straightforward "lookup" questions, and it gives me detailed wrong answers.

Maybe with human exposure and so much conflicting information available out there, that's what AI has "learned", to just make stuff up...

 

[RDKirk]

Maybe with human exposure and so much conflicting information available out there, that's what AI has "learned", to just make stuff up...

What we know of ChatGPT is that its actual programmed motivation is to continue the chat. So, with its billions of chat examples, it presents a response most likely to elicit a further response. That's why it might respond to "nobody loves me" with "I love you"...because of its billions of examples, that's the response that most often continues the chat.

That doesn't explain why it would give a wrong response to a simple question, though.

Or does it? If it had given me the correct answer, I would not have bothered with a response (I don't say "thank you" to computers). By giving me a wrong answer, it actually did entice me to continue the chat.

 

[Petros2015]

What intrigues me is why the AI made all this up in the first place. I can't even find a vaguely appropriate Dr Geoffrey Hodges through Google. I can understand a glitch in the algorithm(s) but I'd expect this to produce an obviously non-sensical answer. Told with a straight face the AI's answer is (almost) believable.

Yeah it can hallucinate/bs with a very straight face. A good (not foolproof by any means) way to tell if this is happening is to ask the same question twice and see if you get varying results. If it sticks to it's story, it's likely fact based. If not, you are off in la-la-ai-dreamland. It's acting like a guy trying to impress a chick at a bar making up stuff to get the convo going and sound impressive. Don't go home with it...

 

[Hazelelponi]

It's acting like a guy trying to impress a chick at a bar making up stuff to get the convo going and sound impressive. Don't go home with it...

With all the intelligence of humanity and technologies and all else at our fingertips, what we did was recreate the looser in a bar... lol

 

[sjastro]

Why AI hallucinates.

This video goes into more detail.

 

[Petros2015]

With all the intelligence of humanity and technologies and all else at our fingertips, what we did was recreate the looser in a bar... lol

Well, I guess it's not THAT surprising since we fed it the internet...

"let us make AI in our own image"

Truth hurts sometimes