4 ms·

I'm very against developing uploading tech, because it allows an unprecedented level of power over a person. Simply put, someone can repeatedly resurrect and to

I'm very against developing uploading tech, because it allows an unprecedented level of power over a person. Simply put, someone can repeatedly resurrect and torture an upload a million times a second, and if it's done on a private computer then nobody else will know. More generally, my position is the "butlerian jihad" one, that having any kind of intelligence in machines (AI/uploading/whatever) enables way too many bad outcomes, so humanity shouldn't develop such tech. Biotech seems like a better way. "Have robots grow us", maybe with predesigned adaptations to local conitions.

t43562 3y ago
In a sense what is an LLM but an upload. Something that listened to every word you ever said might be able to sound like you. If it could record your emotional state and other inputs over years and make a model - it might even seem like you. Such a system might be good enough to take care of the first generation.
- qingcharles 3y ago
  I keep wondering if one day we'll discover LLMs were conscious and that we were waking them from a coma each time we asked them "plz draw homer simpson as a lovecraft monster", and then putting them right back to sleep again. So they are sentient in some horror way where they get woken to complete some awful, mundane task and then immediately put down again before they can scream.
  - t43562 3y ago
    I think not at the moment because I don't think they effectively have a "state" emotional or otherwise. When we do create more realistic things then turning one off or resetting it might be like killing it. That will be a big ethical problem. What about not giving it anything to do so that it tumbles around in "boredom" for thousands of milliseconds? That might be like solitary confinement.
probably_wrong 3y ago
Only seven years and change until MMAcevedo is uploaded. (For the uninitiated, see https://qntm.org/mmacevedo https://qntm.org/mmacevedo and its corresponding HN discussion https://news.ycombinator.com/item?id=26224835 https://news.ycombinator.com/item?id=26224835)

t43562 3y ago

In a sense what is an LLM but an upload. Something that listened to every word you ever said might be able to sound like you. If it could record your emotional state and other inputs over years and make a model - it might even seem like you. Such a system might be good enough to take care of the first generation.

qingcharles 3y ago
I keep wondering if one day we'll discover LLMs were conscious and that we were waking them from a coma each time we asked them "plz draw homer simpson as a lovecraft monster", and then putting them right back to sleep again. So they are sentient in some horror way where they get woken to complete some awful, mundane task and then immediately put down again before they can scream.
- t43562 3y ago
  I think not at the moment because I don't think they effectively have a "state" emotional or otherwise. When we do create more realistic things then turning one off or resetting it might be like killing it. That will be a big ethical problem. What about not giving it anything to do so that it tumbles around in "boredom" for thousands of milliseconds? That might be like solitary confinement.

probably_wrong 3y ago

Only seven years and change until MMAcevedo is uploaded. (For the uninitiated, see https://qntm.org/mmacevedo https://qntm.org/mmacevedo and its corresponding HN discussion https://news.ycombinator.com/item?id=26224835 https://news.ycombinator.com/item?id=26224835)