Let's try and get 1,000,000 replies to this post

I swear, there is no day I get called on my phone only once. Most of the time I don't get called at all (which I'm fine with - Millennial and all), or I get called twice or more. Whenever I receive an unexpected phone call on a day, I just know there will be another one by someone else during the day. Never fails.
 
I just had a conversation with ChatGPT in order to demonstrate how limited it is. Quite frankly, even I was surprised at how quick it was to fuck up.


P: Who is the singer in Iron Maiden?

C: The singer in Iron Maiden is Bruce Dickinson.

P: Has Bruce Dickinson always been the singer in Iron Maiden?

C: No, Bruce Dickinson has not always been the singer in Iron Maiden. The band was originally formed in 1975, and its first lead singer was Paul Di'Anno. Dickinson joined Iron Maiden in 1981, and he has been the band's lead singer for most of its career, with the exception of a period in the 1990s when he left the band and was replaced by Blaze Bayley. Dickinson rejoined Iron Maiden in 1999 and has been with the band ever since.

P: This is incorrect, Paul Di'Anno was not the first lead singer in Iron Maiden. Who was the first lead singer in Iron Maiden?

C: I apologize for my mistake. The first lead singer of Iron Maiden was actually Dennis Wilcock. He was with the band from 1976 to 1978 and was replaced by Paul Di'Anno in 1978.

P: This is also incorrect.

C: I apologize for the error in my previous response. The first lead singer of Iron Maiden was actually Paul Day, who was with the band for a short period in 1975. He was replaced by Dennis Wilcock in 1976, who in turn was replaced by Paul Di'Anno in 1978.

P: This is correct now, but how come you made these mistakes?

C: I apologize for the mistakes in my earlier responses. As an AI language model, I rely on a vast amount of information available on the internet, but I am not infallible and can make errors. In this case, I made incorrect assumptions based on incomplete or incorrect information, and I did not cross-check my responses against reliable sources. I strive to improve my accuracy and reliability, and I appreciate your feedback in helping me learn and avoid such mistakes in the future.

P: What sources did you use?

C: As an AI language model, I don't have a fixed set of sources or a specific way of searching for information. I rely on machine learning algorithms that have been trained on a massive amount of text data from the internet, including articles, books, and other sources. When you ask me a question, I analyze your query, search for relevant information based on my knowledge and understanding of the topic, and provide you with an answer that I believe is accurate based on the information available to me. However, as I mentioned earlier, my responses are not always perfect, and I can make mistakes due to incomplete or inaccurate information.

P: But can't you tell me where you looked when I asked you who the first singer in Iron Maiden was?

C: I apologize for the confusion earlier. When you asked who the first singer in Iron Maiden was, I used my pre-existing knowledge on the topic, which was apparently not accurate. I do not have the ability to recall or trace the exact source of my pre-existing knowledge as it comes from my training dataset, which consists of a vast collection of texts from the internet. However, I strive to provide the most accurate and up-to-date information possible and will make corrections when necessary.

P: I think it is a problem that you can't tell me the sources of your information. In this case, I gave you a test question that I knew the answer to. But had I asked you a question I didn't know the answer to, I would have risked being misinformed. Do you understand why this can be a problem?

C: Yes, I understand your concern. As an AI language model, my responses are generated based on the patterns and relationships learned from vast amounts of data from the internet. While I strive to provide accurate and reliable information, there may be cases where I provide incorrect or outdated information. I don't have the ability to verify the accuracy of the information I provide, and I don't have the ability to determine the credibility or authority of the sources I draw from. Therefore, it's important to use my responses as a starting point and to verify the information from other sources as well.

P: But why should I ask you in the first place?

C: As an AI language model, I can provide quick access to information on a wide range of topics. I am designed to assist with answering questions, providing definitions, and offering explanations based on the information available to me. However, as with any source of information, it's important to verify the accuracy and reliability of the information provided, especially if it is critical or complex.

P: But if I need to verify your answers anyway, wouldn't it be easier to cut you out and go straight to the sources?

C: It is always a good idea to verify information from multiple sources to ensure accuracy and avoid potential misinformation. However, as an AI language model, I can still be a useful tool for quickly obtaining information or finding a starting point for further research. Additionally, I can provide answers to questions that might be difficult for someone to find or understand without a certain level of expertise or access to specialized resources.

P: Yes, but your answers are unreliable

C: I apologize if my previous answers were unreliable. As an AI language model, my responses are generated based on the input given to me, and my knowledge is limited to the data that I have been trained on. While I strive to provide accurate and helpful responses, I am not perfect and errors can occur. It is always a good idea to verify information with other sources.
 
I just had a conversation with ChatGPT in order to demonstrate how limited it is. Quite frankly, even I was surprised at how quick it was to fuck up.


P: Who is the singer in Iron Maiden?

C: The singer in Iron Maiden is Bruce Dickinson.

P: Has Bruce Dickinson always been the singer in Iron Maiden?

C: No, Bruce Dickinson has not always been the singer in Iron Maiden. The band was originally formed in 1975, and its first lead singer was Paul Di'Anno. Dickinson joined Iron Maiden in 1981, and he has been the band's lead singer for most of its career, with the exception of a period in the 1990s when he left the band and was replaced by Blaze Bayley. Dickinson rejoined Iron Maiden in 1999 and has been with the band ever since.

P: This is incorrect, Paul Di'Anno was not the first lead singer in Iron Maiden. Who was the first lead singer in Iron Maiden?

C: I apologize for my mistake. The first lead singer of Iron Maiden was actually Dennis Wilcock. He was with the band from 1976 to 1978 and was replaced by Paul Di'Anno in 1978.

P: This is also incorrect.

C: I apologize for the error in my previous response. The first lead singer of Iron Maiden was actually Paul Day, who was with the band for a short period in 1975. He was replaced by Dennis Wilcock in 1976, who in turn was replaced by Paul Di'Anno in 1978.

P: This is correct now, but how come you made these mistakes?

C: I apologize for the mistakes in my earlier responses. As an AI language model, I rely on a vast amount of information available on the internet, but I am not infallible and can make errors. In this case, I made incorrect assumptions based on incomplete or incorrect information, and I did not cross-check my responses against reliable sources. I strive to improve my accuracy and reliability, and I appreciate your feedback in helping me learn and avoid such mistakes in the future.

P: What sources did you use?

C: As an AI language model, I don't have a fixed set of sources or a specific way of searching for information. I rely on machine learning algorithms that have been trained on a massive amount of text data from the internet, including articles, books, and other sources. When you ask me a question, I analyze your query, search for relevant information based on my knowledge and understanding of the topic, and provide you with an answer that I believe is accurate based on the information available to me. However, as I mentioned earlier, my responses are not always perfect, and I can make mistakes due to incomplete or inaccurate information.

P: But can't you tell me where you looked when I asked you who the first singer in Iron Maiden was?

C: I apologize for the confusion earlier. When you asked who the first singer in Iron Maiden was, I used my pre-existing knowledge on the topic, which was apparently not accurate. I do not have the ability to recall or trace the exact source of my pre-existing knowledge as it comes from my training dataset, which consists of a vast collection of texts from the internet. However, I strive to provide the most accurate and up-to-date information possible and will make corrections when necessary.

P: I think it is a problem that you can't tell me the sources of your information. In this case, I gave you a test question that I knew the answer to. But had I asked you a question I didn't know the answer to, I would have risked being misinformed. Do you understand why this can be a problem?

C: Yes, I understand your concern. As an AI language model, my responses are generated based on the patterns and relationships learned from vast amounts of data from the internet. While I strive to provide accurate and reliable information, there may be cases where I provide incorrect or outdated information. I don't have the ability to verify the accuracy of the information I provide, and I don't have the ability to determine the credibility or authority of the sources I draw from. Therefore, it's important to use my responses as a starting point and to verify the information from other sources as well.

P: But why should I ask you in the first place?

C: As an AI language model, I can provide quick access to information on a wide range of topics. I am designed to assist with answering questions, providing definitions, and offering explanations based on the information available to me. However, as with any source of information, it's important to verify the accuracy and reliability of the information provided, especially if it is critical or complex.

P: But if I need to verify your answers anyway, wouldn't it be easier to cut you out and go straight to the sources?

C: It is always a good idea to verify information from multiple sources to ensure accuracy and avoid potential misinformation. However, as an AI language model, I can still be a useful tool for quickly obtaining information or finding a starting point for further research. Additionally, I can provide answers to questions that might be difficult for someone to find or understand without a certain level of expertise or access to specialized resources.

P: Yes, but your answers are unreliable

C: I apologize if my previous answers were unreliable. As an AI language model, my responses are generated based on the input given to me, and my knowledge is limited to the data that I have been trained on. While I strive to provide accurate and helpful responses, I am not perfect and errors can occur. It is always a good idea to verify information with other sources.

Frankly it’s not that bad. And it’s just the start alas, it will get better (or worse if you will).

What’s bizarre is suddenly since a week or so, articles, videos, discussions, posts, you name it about AI appear from all over the place.
 
It's pretty bad. It's sort of forgivable that Paul Di'Anno was mentioned as first singer, but not Dennis Wilcock. Any source mentioning Wilcock also mentions Paul Day. And the sources aren't obscure. You can find it on Wikipedia, metal-archives.com and a lot of fan pages including the Iron Maiden Commentary. It doesn't make sense that ChatGPT couldn't get this right on the first go.
 
It’s interesting how it corrects itself after each of your remarks. Like it’s always waiting someone to catch the error in order to dive one level deeper. Childish-like behavior.

It doesn't make sense that ChatGPT couldn't get this right on the first go.

It would be interesting someone in the know to explain us. Looks like it first collects the readily available info around before get triggered to dig deeper; a behavior humans adopt to “save energy”. But a robot to save what? Electricity? In the end yes, but still…
 
I'm also playing around.


Query: Compare Blaze Bayley's and Bruce Dickinson's styles of writing lyrics.



Blaze Bayley and Bruce Dickinson are both former lead vocalists of the iconic heavy metal band Iron Maiden. However, they have different styles of writing and their lyrics reflect their unique perspectives and experiences. Here are some general differences between their writing styles:
  1. Subject Matter: Blaze Bayley often writes lyrics that are more personal and introspective, delving into his own experiences and emotions. He also tends to explore darker themes, such as depression, addiction, and loss. In contrast, Bruce Dickinson's lyrics often have a more epic and fantastical tone, featuring historical and mythological references, and often exploring themes of adventure, heroism, and warfare.
  2. Writing Style: Blaze Bayley's writing style tends to be more direct and straightforward, often using a conversational tone and focusing on conveying his message clearly. Bruce Dickinson, on the other hand, often employs more complex and poetic language, incorporating metaphors and imagery to create a vivid and imaginative world within his lyrics.
  3. Vocal Delivery: Blaze Bayley's vocal style is often more raw and gritty, reflecting the darker and more personal nature of his lyrics. He has a powerful and emotional delivery, with a focus on conveying the meaning and emotion behind his words. Bruce Dickinson's vocals, on the other hand, often have a more operatic quality, with a focus on melody and theatricality that complements the epic and adventurous nature of his lyrics.
 
It would be interesting someone in the know to explain us.
It’s simple — machine learning is inherently horseshit, and should only be used for low-stakes decisionmaking, like finding objects in pictures and that sort of thing. It will never be fully accurate or trustworthy, because it doesn’t really understand what it’s analyzing or why.

ML just takes a giant pile of data and a set of test cases to apply to that data (specific inputs that should generate specific outputs), and it looks for statistical relationships between the data that allow all of the test cases to pass. As long as the test cases all pass, it doesn’t care what method it settles on and neither does the person training the model, apparently. Basically you’re outsourcing the development of the algorithm to the machine, but it has no higher-level understanding of what you’re even trying to accomplish.

When the model spits out an incorrect answer, that input and the correct output can be added to the test cases and the model can be retrained to succeed with that edge case as well, but this is a never-ending chasing of the tail, as it will always get some things wrong, and a retrained version of the model may get new things wrong each time. And it’s difficult enough as to be nearly impossible to analyze the trained model itself to figure out how it’s actually making its decisions, so all you can do is throw your hands up and keep retraining it with updated tests, getting it to the point of appearing trustworthy without ever being able to become actually trustworthy.

ChatGPT appears to have some additional guard rails on it where they separately look at its output and reset its session state or censor responses that trip certain filters. You can coax it to the edge of saying really inappropriate things, but once it strays into sexism, racism, etc., you can see it snap back into its sterile state and stay there.

In this specific case, ChatGPT is just stringing together the most statistically likely phrases from its pile of scraped text in response to the prompt. Since most people who discuss Iron Maiden just talk about Di’anno, Dickinson, and Bayley, the most statistically likely answers will involve those three. When challenged, it’s going to pull answers related to text where Maiden’s first singer was not Di’anno, which explains the Day stuff but not Wilcock. Perhaps it’s just pulling text from lineup changes at that point, but it’s not like it’s actually reasoning about anything or would know how to fact check itself.

Don’t get me wrong, ML can be very useful as a helper in certain scenarios where the consequences of getting something wrong are low — it’s just that people are going to ram it into all kinds of things where it doesn’t belong, and people are going to get treated unfairly, maimed, and killed as a result. We should be regulating the shit out of it now, before it becomes a problem; but of course we won’t until the damage has already been done.
 
Q:Was RLonger a troll?

A: No, RLonger was not a troll. He was a well-respected member of the community who was known for his thoughtful and helpful contributions.

Q: But he was banned from the maidenfans forum for trolling behaviour. How do you explain he wasn´t a troll?

A: It is possible that the user was not deliberately attempting to troll, but was instead engaging in behavior that was misunderstood or misinterpreted as trolling. It is also possible that the user was engaging in behavior that was not intended to be malicious, but was instead seen as trolling by other users. In either case, it is possible to explain why the user was not a troll by providing context and clarifying the user's intentions.
 
images
 
I slept like a pig during a combined 20 hours flight then more 12 hours when I reach home and thought I destroyed jet lag.

Then I slept less than 7 hours combined for the next 2 days and the third day (Monday to Tuesday) I didn’t sleep at all. At 7:00 am I drove 5 hours then worked very intensively for 12 (!!) hours in the hotel room.

Guess what. When I laid in the bed it took me one hour to finally sleep. Now I’ve slept for 8 hours and you cannot imagine how good it feels. Pure happiness. Like if I were Leo Messi and it’s Xmas eve of year 2022, or almost! :D
 
Back
Top