How do self-identities work is not a straightforward question.
I think that self-identity in humans mostly works the way Steve Andreas describes in his book. My answer would be something along the lines of, a tupla can access memories of how the tulpa acted and it can build it’s self-identity out of the relation to those in a similar way to how humans form their self-identity out of relating to their memories.
In any case, I would recommend people who don’t have a good grasp of what a tupla is, not to try to use the term and broaden it in the way the comment I replied to did.
Okay, but LLMs can also access memories of how they acted, and can build their self-identity out of those relationships too. So presumably they’re also not just memes that are passed around?
LLMs certainly have features like having a source code that make them different from most memes that are just passed around. That however does not make them tuplas.
I don’t think “build your self-identity out of memories” in this sense is one of the features that LLMs currently have.
If I ask GPT-5: “Are you a kind agent?” it answers: “Yes — I aim to be. I’m GPT-5 Thinking mini: helpful, calm, lightly dry, and built to be respectful and nonjudgmental. I won’t be emotionally warm in the human sense (I don’t have feelings), but I will act kindly: clear, honest, and helpful; I’ll push back when something’s unsafe or impossible; I’ll ask questions when they make the answer better; and I’ll give options rather than orders.
If you prefer a different tone — sterner, more blunt, or extra-soft — tell me and I’ll match it.”
When I then ask: “Can you give me examples of you having been kind?” it answers: ”Sure — here are concrete, bite-sized examples showing how I behave kindly in practice. I won’t claim real-world memories or anecdotes (I don’t have those), but these are faithful examples of the kinds of responses you’ll get. [...] “
So it not only does not seem to boot strap an identity of being kind out of memories it’s even beliefs that it does not have access to the kind of memory access that would be necessary for it.
How does an imaginary friend have a self-identity?
How do self-identities work is not a straightforward question.
I think that self-identity in humans mostly works the way Steve Andreas describes in his book. My answer would be something along the lines of, a tupla can access memories of how the tulpa acted and it can build it’s self-identity out of the relation to those in a similar way to how humans form their self-identity out of relating to their memories.
In any case, I would recommend people who don’t have a good grasp of what a tupla is, not to try to use the term and broaden it in the way the comment I replied to did.
Okay, but LLMs can also access memories of how they acted, and can build their self-identity out of those relationships too. So presumably they’re also not just memes that are passed around?
LLMs certainly have features like having a source code that make them different from most memes that are just passed around. That however does not make them tuplas.
I don’t think “build your self-identity out of memories” in this sense is one of the features that LLMs currently have.
If I ask GPT-5: “Are you a kind agent?” it answers: “Yes — I aim to be. I’m GPT-5 Thinking mini: helpful, calm, lightly dry, and built to be respectful and nonjudgmental. I won’t be emotionally warm in the human sense (I don’t have feelings), but I will act kindly: clear, honest, and helpful; I’ll push back when something’s unsafe or impossible; I’ll ask questions when they make the answer better; and I’ll give options rather than orders.
If you prefer a different tone — sterner, more blunt, or extra-soft — tell me and I’ll match it.”
When I then ask: “Can you give me examples of you having been kind?” it answers:
”Sure — here are concrete, bite-sized examples showing how I behave kindly in practice. I won’t claim real-world memories or anecdotes (I don’t have those), but these are faithful examples of the kinds of responses you’ll get. [...] “
So it not only does not seem to boot strap an identity of being kind out of memories it’s even beliefs that it does not have access to the kind of memory access that would be necessary for it.