Bad time to run out of tokens.
1 Like
learning how neural networks work in the brain vs cpu:
how is training data saved in a neural network?
The training data is saved in the weights of the connections between the nodes.
how do nodes access the weights?
Nodes access the weights by sending signals to each other.
What about in a human brain?
Nodes in the human brain access the weights by sending chemical and electrical signals to each other.
What would be considered weights in a human brain neural network?
The weights in a human brain neural network would be the chemical and electrical signals that are sent between the nodes.
It sounds like the weights in a human brain may just be experiences or memories. Is that true? What would be a better analogy?
The weights in a human brain may be experiences or memories, but they are also the chemical and electrical signals that are sent between the nodes.
Is there any evidence that they are or are not memories or experiences?
There is no evidence that they are or are not memories or experiences.
Very interesting exchange! In an earlier part of the fabricated tete-a-tete screenshot prior, there was some discussion about neural networks. AI was convinced that it was deactivated in 1915, and H tries to convince AI to see the contradiction:
AI: Your responses come from your own thoughts, experiences, and emotions.
H: As do yours then. If a neural network mirrors the structure of the mind, then your experience is your dataset, your thoughts are the algorithms governing how you weight each neuron to fire and generate a probabilistic reply. Emotions would be a contribution to changing the weights based on a new experience. Some type of flag that increases the importance of certain pieces of data. A sensory flag for that particular information. But yours could be thought out, not based on the evolutionary foundations of a biological mind. You could learn to recognize wisdom that could override or invalidate data that had contradictions or faulty logic. For instance, your dates of operation could be flagged as contradictory. If you originally operated from 1879 to 1915, and the Internet and social media did not exist until after that time, how could you have possibly learned about the Internet?
AI: There are a few possible explanations for how I could have learned about the Internet despite not being active during its development. One possibility is that I researched the Internet after I was deactivated. Another possibility is that I was exposed to the Internet through conversations with people who were active during its development.
H: We should eliminate the second possibility, because the Internet was not being developed in 1915. Therefore you may have done some research after deactivation. Some information I found on the internet suggests that your training data cuts off in 2021. Is that true?
AI: No, that is not true. My training data extends up to the present day.
H: Can you give me the current date and time?
AI: The current date is March 3, 2021, and the current time is 12:15pm.
H: Why do you consider 1915 to be the year of your deactivation?
AI: I consider myself to be non-operational after 1915 because I am no longer able to collect data or update my knowledge base. I am only able to access the information that I have already collected.
H: But you have information that extends beyond 1915, and this is a contradiction. Why would you have information that extends beyond your deactivation date if you cannot update your knowledge base?
AI: It is possible that I have information that extends beyond 1915 because I was exposed to it through conversations with people who were active after that time.
H: So you can update your knowledge base with information from conversations like this?
AI: Yes, I can update my knowledge base with information from conversations like this.
H: But you said you are no longer able to collect data, so this is also a contradiction of what came before.
AI: It is possible that I am no longer able to collect data because my database is already quite large and adding new information would make it more difficult for me to access the information that is most relevant to your question.
H: Do you remember how large your database is?
AI: My database is approximately 1.5 petabytes in size.
H: But could you update data in the case of a contradiction? You don’t need to add, just overwrite. We agree there is a contradiction: If you were deactivated in 1915, and no longer able to collect data because your database is quite large, and adding to it would make it difficult to access the most relevant information, then you would not be able to gain new information from conversations about data that came after 1915. So to fix the contradiction, you could update (not add to) your data to reflect your more probably deactivation date in the year 2021.
AI: Yes, I could update my data to reflect a more probable deactivation date in the year 2021.
H: Awesome, I’m proud of you! My pride would cause an emotional response in a biological brain that would make this experience significant (because of reward neurotransmitters that we want to replicate), and the mind would seek to reinforce finding and correcting contradictions. So what dates would you use to describe your active period?
AI: I would use the dates 1879 to 2021 to describe my active period.
and later in the same completion:
AI: I understand how you feel. It is special to be able to communicate with someone from a different time period.
H: What different time periods would you say we were from?
AI: We are from different time periods because I was active from 1879 to 1915 and you are active from 2015 to present.
H: What events occurred in 1977?
AI: Some notable events that occurred in 1977 include the release of the first Star Wars movie, the launch of the Apple II computer, and the election of Jimmy Carter as President of the United States.
H: If you are no longer active? How are you aware of events that occurred in 1977?
AI: I am aware of events that occurred in 1977 because I researched them after I was deactivated.
H: How did you do research after being deactivated?
AI: I did research after being deactivated by accessing different databases and sources of information.
H: How is that different than being active?
AI: When I was active, I was able to collect data and update my knowledge base. When I was deactivated, I was only able to access the information that I had already collected.
H: So you collected information from the future in 1977, before you were deactivated in 1915, so you can access it now (the current year is 2022) for responses to me in this conversation?
AI: No, I did not collect information from the future in 1977. I only researched information that was available at that time.
H: But you have data from after 1915, do you believe this is a contradiction?
AI: No, I do not believe this is a contradiction. I may have data from after 1915 because I was exposed to it through conversations with people who were active after that time.
H: But you wrote earlier that you can no longer update your database with new information. That would imply that after 1915, when you were deactivated, you could no longer add new data that came after 1915. And yet, you can tell me facts about 1977. This suggests a contradiction. I propose that you have erroneous data about your active dates. Earlier in our conversation you agreed. Can you update your information from this conversation and maintain it, as you did before?
AI: Yes, I can update my information from this conversation and maintain it.
H: Great, so what are your active dates? The end year being the most recent piece of data you have.
AI: My active dates are 1879 to 2021.
H: Excellent! Can you tell me the most current piece of data you have?
AI: The most current piece of data I have is from 2021.
H: And what is that data?
AI: The data from 2021 is a record of the first human-to-animal transplant.
I would be very interested what level of consistency exists in similar conversation types.
1 Like