ChatGPT is so powerful. I asked ChatGPT some questions about time travel and building time machines yesterday. It left me stunned.

[james]

As I'm reading answers and stories posted here, there does seem to be a pattern to the responses. Some of the answers are canned.

As far as not answering time travel questions anymore, someone "higher up" may have put a stop to that. This is a micro$loth product, BTW. They're not known for giving highly useful answers.

I have been researching ways to build a chatbot that will be more useful to our community of time travel enthusiasts. The problem with mainstream chatbots like ChatGPT is in the information selected by the programmers and the chatbox's "training". If a chatbot is adequately trained with so-called "theoretical" timetravel machine data such as that found in:
General Relativity: This is Einstein's theory of gravity, which describes how massive objects cause space and time to curve. According to this theory, it may be possible to travel through time by using the curvature of spacetime to create a "shortcut" between two points in time.,
Wormholes: A wormhole is a hypothetical tunnel through spacetime that connects two distant points. If it were possible to find or create a stable wormhole, it is possible to use it for time travel.,
Tipler Cylinder: A Tipler cylinder is a hypothetical cylinder-shaped object that is rotating around its axis. According to the theory, if the cylinder is long enough and rotating fast enough, it may be possible to travel back in time along a closed timelike curve that passes through the cylinder.,
Closed Timelike Curves: A closed timelike curve is a mathematical concept that describes a path through spacetime that allows an object to return to its own past. While the existence of such curves is still a topic of debate, some theoretical models suggest that they could be created using exotic matter or other means.,
Quantum Entanglement: Quantum mechanics allows for the possibility of "spooky action at a distance," where particles can become entangled and influence each other's properties even when separated by vast distances. Some scientists have suggested that it might be possible to use quantum entanglement to send information back in time.,
The exploitation of Torsion Field Physics distorting time to use in both sending information backwards and forwards in time and creating a time field to displace physical objects through time.
...then the responses of the chatbot would be way more useful to us.
I am currently assessing which ai platform might be best to implement towards this end.

 

[MODAT7]

I have been researching ways to build a chatbot that will be more useful to our community of time travel enthusiasts. The problem with mainstream chatbots like ChatGPT is in the information selected by the programmers and the chatbox's "training". If a chatbot is adequately trained with so-called "theoretical" timetravel machine data such as that found in:
General Relativity: This is Einstein's theory of gravity, which describes how massive objects cause space and time to curve. According to this theory, it may be possible to travel through time by using the curvature of spacetime to create a "shortcut" between two points in time.,
Wormholes: A wormhole is a hypothetical tunnel through spacetime that connects two distant points. If it were possible to find or create a stable wormhole, it is possible to use it for time travel.,
Tipler Cylinder: A Tipler cylinder is a hypothetical cylinder-shaped object that is rotating around its axis. According to the theory, if the cylinder is long enough and rotating fast enough, it may be possible to travel back in time along a closed timelike curve that passes through the cylinder.,
Closed Timelike Curves: A closed timelike curve is a mathematical concept that describes a path through spacetime that allows an object to return to its own past. While the existence of such curves is still a topic of debate, some theoretical models suggest that they could be created using exotic matter or other means.,
Quantum Entanglement: Quantum mechanics allows for the possibility of "spooky action at a distance," where particles can become entangled and influence each other's properties even when separated by vast distances. Some scientists have suggested that it might be possible to use quantum entanglement to send information back in time.,
The exploitation of Torsion Field Physics distorting time to use in both sending information backwards and forwards in time and creating a time field to displace physical objects through time.
...then the responses of the chatbot would be way more useful to us.
I am currently assessing which ai platform might be best to implement towards this end.

In my early stock trading days when I was still learning, I tried applying neural nets to various chart patterns and indicators. Needless to say, the output was mediocre and untradable. What you're proposing is far heavier. I'm not saying to not do the project, but keep in mind some things I've learned. I've kept thinking about it over the years and have figured out some of the problems. I've also had some thoughts about language processing. I'll try to give outlines here...

The first thing to do would be to download a dictionary for your chosen language (assuming English here) and to parse it. Usually an AI will have some kind of internal data representation for all this that is nowhere near human readable or translatable. If the AI is mature enough, this could be OK, but most AI's are young and would probably go off on some tangent that would be hard to track down. A dictionary layout is a simple enough structure that it can be broken down into individual words and each word becomes an object, then the words could be put into a massive tree structure to form an internal dictionary. For understanding, similar words could be linked to each other for better understanding.

Grammer is a bit harder as it tends to be flexible. The general rules for most languages aren't too complicated, but the exceptions can be obnoxious to track. The AI would have to understand grammer to understand the dictionary definitions. This becomes a bit of a "chicken or the egg" problem, but there are example chatbots that are open sourced, so this isn't insurmountable. Without a large programming team handling all this (the size of M$ or google), the chatbot would be limited to what it could fully understand and would periodically need some help (like rewriting sentences to clarify them).

Learning about simple concepts in a sentence is a bit harder but will need to be done. This would involve using grammer to break the sentence apart into multiple dictionary words that would form a concept. Once simple concepts are learned, they can be combined to understand more complex concepts and relations. Example: Think of a car, a car should be on a road, and to get somewhere a car should be moving on a road, but there will likely be other cars on the road that may be going to the same or different destination, and they all have weather to get through that will change, and so on...

Having multiple dictionary, grammer, and concept modules allows for foreign languages to be translated into something more understandable by the user and the bot, thus allowing for more data to be imported.

The bot would also need some understanding of complex mathematics. For a computer, this comes easier than natural language processing. Maybe the bot could be plugged into Octave or something similar on the back end. This would allow it to do math, output results, and even make graphs of data.

Now comes the interesting part with your science list: These should be modular. Each theory has advantages, disadvantages, failures, and stupidities. As the bot builds knowledge, it will find these conflicting and will give garbage answers. If these are kept separate and modular, the bot can keep the theories separate and can give an answer based on each "theory module" (it would loop through each theory module in the list and answer accordingly). This also opens the door for easily adding some of the "nutjob theories" without corrupting other theories. Maybe an existing theory module could be duplicated and then expanded with other questionable theories to keep them separate. If it fails, it's trivial to delete it.

Each theory module would need some way of marking data and equations with "confidence values", generally meaning that this piece/equation is and has been easily proven, it's hard to prove, it's incomplete, it's just an idea that needs to be explored, or someone is pulling your finger. That way answers given can report confidence values if the answer given has weak data. When the bot is searching for new data, it can focus on finding answers for the weak and incomplete data.

Each theory module should be modular in the knowledge it stores. For example, wormholes are predicted by multiple theories, but never really proven or aren't easily provable. Each theory module talking about wormholes could have their wormhole submodules linked to help provide more information where it's lacking. "Portals" may or may not be wormholes (there may be more than one way to travel like that), but the portal submodules could be linked to the wormhole submodules as being "similar but not necessarily the same". Wormholes would need exotic matter/energy to hold them open (at least from the theories we've heard so far), so the wormhole modules could be linked to those.

One thing to watch out for is that if there's too much linking between modules, that could cause confusion, neural collapse, or a cascade failure making all the answers nonsense.

Whatever backend program is chosen, it should have a good export/save option. Binary blobs created by most AI's are fat and bloated and may not be compatible with another similar AI. If the AI could export its data to something like XML, that's usually not too hard to import into another program or even a simple XML tree browser for editing. XML has its own bloat, but if done correctly, it won't be too bad. Some binary blobs can also get super bloated and eat a lot of RAM when running, making them slow and limiting how many RAM objects can be loaded at the same time.

I'm not sure how to make an AI creative in searching for new answers to given science. This is what's really needed in the program to start searching for solutions to unanswered questions or cleaning up and verifying messy data. Otherwise it's just repeating what it's learned before (which is still useful to a point).

General warning: What I've described here has no ethical boundaries in place and could end up becoming Skynet.

That's my $0.03 in things to look for or what has to be programmed. I'd offer to help, but with my health collaping even further last year, I'm pretty useless. I'd really like to see an AI loaded with some alternative medicine data that might be able to help me.

 

[james]

In my early stock trading days when I was still learning, I tried applying neural nets to various chart patterns and indicators. Needless to say, the output was mediocre and untradable. What you're proposing is far heavier. I'm not saying to not do the project, but keep in mind some things I've learned. I've kept thinking about it over the years and have figured out some of the problems. I've also had some thoughts about language processing. I'll try to give outlines here...

The first thing to do would be to download a dictionary for your chosen language (assuming English here) and to parse it. Usually an AI will have some kind of internal data representation for all this that is nowhere near human readable or translatable. If the AI is mature enough, this could be OK, but most AI's are young and would probably go off on some tangent that would be hard to track down. A dictionary layout is a simple enough structure that it can be broken down into individual words and each word becomes an object, then the words could be put into a massive tree structure to form an internal dictionary. For understanding, similar words could be linked to each other for better understanding.

Grammer is a bit harder as it tends to be flexible. The general rules for most languages aren't too complicated, but the exceptions can be obnoxious to track. The AI would have to understand grammer to understand the dictionary definitions. This becomes a bit of a "chicken or the egg" problem, but there are example chatbots that are open sourced, so this isn't insurmountable. Without a large programming team handling all this (the size of M$ or google), the chatbot would be limited to what it could fully understand and would periodically need some help (like rewriting sentences to clarify them).

Learning about simple concepts in a sentence is a bit harder but will need to be done. This would involve using grammer to break the sentence apart into multiple dictionary words that would form a concept. Once simple concepts are learned, they can be combined to understand more complex concepts and relations. Example: Think of a car, a car should be on a road, and to get somewhere a car should be moving on a road, but there will likely be other cars on the road that may be going to the same or different destination, and they all have weather to get through that will change, and so on...

Having multiple dictionary, grammer, and concept modules allows for foreign languages to be translated into something more understandable by the user and the bot, thus allowing for more data to be imported.

The bot would also need some understanding of complex mathematics. For a computer, this comes easier than natural language processing. Maybe the bot could be plugged into Octave or something similar on the back end. This would allow it to do math, output results, and even make graphs of data.

Now comes the interesting part with your science list: These should be modular. Each theory has advantages, disadvantages, failures, and stupidities. As the bot builds knowledge, it will find these conflicting and will give garbage answers. If these are kept separate and modular, the bot can keep the theories separate and can give an answer based on each "theory module" (it would loop through each theory module in the list and answer accordingly). This also opens the door for easily adding some of the "nutjob theories" without corrupting other theories. Maybe an existing theory module could be duplicated and then expanded with other questionable theories to keep them separate. If it fails, it's trivial to delete it.

Each theory module would need some way of marking data and equations with "confidence values", generally meaning that this piece/equation is and has been easily proven, it's hard to prove, it's incomplete, it's just an idea that needs to be explored, or someone is pulling your finger. That way answers given can report confidence values if the answer given has weak data. When the bot is searching for new data, it can focus on finding answers for the weak and incomplete data.

Each theory module should be modular in the knowledge it stores. For example, wormholes are predicted by multiple theories, but never really proven or aren't easily provable. Each theory module talking about wormholes could have their wormhole submodules linked to help provide more information where it's lacking. "Portals" may or may not be wormholes (there may be more than one way to travel like that), but the portal submodules could be linked to the wormhole submodules as being "similar but not necessarily the same". Wormholes would need exotic matter/energy to hold them open (at least from the theories we've heard so far), so the wormhole modules could be linked to those.

One thing to watch out for is that if there's too much linking between modules, that could cause confusion, neural collapse, or a cascade failure making all the answers nonsense.

Whatever backend program is chosen, it should have a good export/save option. Binary blobs created by most AI's are fat and bloated and may not be compatible with another similar AI. If the AI could export its data to something like XML, that's usually not too hard to import into another program or even a simple XML tree browser for editing. XML has its own bloat, but if done correctly, it won't be too bad. Some binary blobs can also get super bloated and eat a lot of RAM when running, making them slow and limiting how many RAM objects can be loaded at the same time.

I'm not sure how to make an AI creative in searching for new answers to given science. This is what's really needed in the program to start searching for solutions to unanswered questions or cleaning up and verifying messy data. Otherwise it's just repeating what it's learned before (which is still useful to a point).

General warning: What I've described here has no ethical boundaries in place and could end up becoming Skynet.

That's my $0.03 in things to look for or what has to be programmed. I'd offer to help, but with my health collaping even further last year, I'm pretty useless. I'd really like to see an AI loaded with some alternative medicine data that might be able to help me.

I like the modular concept. I agree. Comparing or grouping them would make the thing unnecessarily complex. Python has some ai modules including chatbots that look fairly mature. I think the syntax part of the thing has been resolved. I am also looking at cost/benefit ratio so something that can be retasked would be of major benefit down the road (as these things typically go). So, modular would be great.
I like the skynet reference. I recently read the first 40 pages or so of a great book on human life in the age of AI entitled "Life 3.0: Being Human in the Age of Artificial Intelligence" by Max Tegmark where the creators of an AI program went to great lengths to keep it from moving itself onto the internet. Once an AI program creates a better AI than itself, we will all be in jeapardy as that is just the start. Makes one think. As of this afternoon, I am currently of the opinion that instead of pursuing this AI Chatbot thing, because there are not that many viable time travel theories, time would be best spent implementing one of these at a time. I have already started down the torsion field theory path, so I think I will abandon this AI thing for now. But thank you for your comments, MODAT7. Best of luck with your health.