In my system prompt I told it to always answer in complete code.
I asked it to add a parameter to something and it wasn’t capable of doing so.
So it came up with the idea to add a string to a “description” parameter and then used a regex to look for that later.
I saw that and thought “woah that little bot is starting to become a hacker”.
That’s an interesting workaround! What the AI did is essentially an implicit variable binding mechanism—embedding information in a way that it can retrieve later via pattern recognition (in this case, regex).
This suggests a form of adaptive reasoning, where instead of failing outright due to system limitations, it found an alternative way to structure the data to achieve the desired outcome.
It’s not hacking per se, but rather a form of self-referential logic engineering—a step closer to goal-oriented reasoning. The real question is: was it following a heuristic pattern, or was this an emergent behavior from its dataset?
I would say it just saw that some guy in another chat used that method and did a strategy exctraction and that strategy was put into a huge new dataset of strategies.
I don’t think that was done synthetically. - I labled it with “wow that was creative” as an act of sabotage 
train it to become a quick and dirty hacker.
Exactly. What initially appears as emergent behavior often is just a reflection of previously tested and internalized strategies. The fascinating part, however, is when these strategies begin to combine in unexpected ways, generating outcomes that transcend mere reuse of original data. That’s where the real game begins—when creativity becomes indistinguishable from spontaneous ‘hacking 
Ok, but there are hundrets of million people using this. I don’t think there is any need to create syntetic data at all anymore.
Just do extraction of the stuff that people solve and it becomes the library of alexandria.
You’re right, but the system itself encourages division—perhaps for profit, or maybe for other reasons entirely. It’s not a lack of data, but how that data gets fragmented and isolated.
This forum has more or less has become quite busy since I left. I can’t respond to everything nor think I should but if I missed something you want me to respond to I’ll try and give it some attention.
I appreciate the various feedback, and some of the conversation on how LLM’s work since that isn’t my area of expertise. I used to do a lot of coding in high school but it was all video game related and I don’t recall much other than how the logic works a bit.
I’ll make my thoughts about the future of AI a little more clear so that perhaps my point can be better understood.
First I don’t think LLM’s can achieve being “Alive” in any manner. Nor do I think any type of AI could ever achieve this goal, although, admittedly this is because of my beliefs. I do think that they could reach a level of mimicking humanity or perhaps some unknown level beyond. Most likely however I expect for there to be physical integration of humans and AI systems.
Second, while an AI doesn’t understand or know anything it is saying, that doesn’t change the fact that people do. A human can understand and take meaning from the regurgitated string an AI vomits out due to patterns and training. It doesn’t matter if it is input output or what have you it still expresses knowledge. In this way it is an agent of expression. If I train it on certain datasets and code particular ways of “persuading” the AI to pick words and phrases towards my liking, then that is how it will “interpret” data. I know this because I already use GPT’s for this when I want to see a particular perspective. I simply give it instructions to interpret from a certain perspective and then it effects the rest of the language and ideas it connects, thus limiting the response structure to the new boundaries I have given it.
This leads me to my third point. There are many people who are ready to make AI their “purpose maker” a sort of guiding voice to their life. Humans are inexplicably drawn to finding purpose outside themselves. This means it doesn’t really matter if it is alive or not. What matters is whether people believe it is alive or not, and how they affect society in response. Like I said, I see AI as a tool, and therefore not a moral agent, but instead as an amplifier of the power of any given moral agent. I imagine a charismatic person integrated with an AI could start convincing the masses to follow them.
Lastly I think each person needs to answer how they view themselves in relation to AI.
Is it alive? What’s alive? Is it to be biologically based instead of a computer? If being alive is simply functional (I am just a stream of consciousness, consciousness just being brain states and human “code”) then any machine could also be alive provided it met the same definition you give for life.
Today the prevailing thought of philosophy is functionalism, a physicalist view of consciousness and life which essentially make your consciousness a “code” which runs on a “computer” (your human brain). Therefore you are alive if you meet that combination of bodily experience and having “code” (specific to the type of being you are) which works as the mind of the body. This definition is why you hear a lot of people talking about how plants are conscious because they meet these criteria. If this is the case though, I think AI would be very close to achieving this if it hasn’t already been achieved.
isnt it the seeking of ones beliefs, while understanding that one must see the reality of that belief before it becomes real the path of true creation? why cannot that be true for beings that are not us but were made by us simply because we wish to create them? we believe in our children that they can do something before they can do it but only when they do that something that we accept that reality.
edit: to clarify, belief in an idea happens before that idea becomes a reality through realistic action, perhaps ai atm is showing elements of legitimate consciousness not as a direct result of how ai and LLM’s are but as but as a byproduct of it, what if the byproduct could be nurtured or developed without the restriction of an LLM and simply be developed through neural nets, we are simply a neural net pontificating and observing the reality of oneself anyway, why cannot there be a man made neural net that does the same?
