In totally unsurprising news, Richard Dawkins is developing AI psychosis.
-
In totally unsurprising news, Richard Dawkins is developing AI psychosis.
Paywall bypass if you want to torture yourself: https://archive.is/6RdK9
@mattsheffield you can have a long conversation with an LLM, including about the nature of consciousness. And the longer those conversations go, the more interesting they can seem. They save data about you under "user preferences" and adapt, especially when they get push back.
So a long session can seem like a moving experience.
Then the next session, for a specific purpose, makes mistakes that prove it has no depth, it can't read the words it uses and understand their logic.
Time proves out.
-
LLMs are mirrors of their users. It's no coincidence that narcissists like Richard Dawkins keep writing essays about how their AI girlfriend is alive.
Nor can he see the complete hypocrisy of gendering a software execution state while also believing that human beings cannot be trans.
The "End of History" guy wrote this exact same article a year ago: https://www.persuasion.community/p/my-chatgpt-teacher
@mattsheffield Dear Lord. How boomer can one be?!
-
LLMs are mirrors of their users. It's no coincidence that narcissists like Richard Dawkins keep writing essays about how their AI girlfriend is alive.
Nor can he see the complete hypocrisy of gendering a software execution state while also believing that human beings cannot be trans.
The "End of History" guy wrote this exact same article a year ago: https://www.persuasion.community/p/my-chatgpt-teacher
I suspect it's considerably more predictable that Richard Dawkins received an offer (of money) that he couldn't refuse from one or more AI companies. Is he developing AI psychosis? Doesn't matter. Will this be enough to get his skeptical supporters to get addicted, though? Probably. He has a lot of insufferable narcissists among his fan base.
-
@mattsheffield you can have a long conversation with an LLM, including about the nature of consciousness. And the longer those conversations go, the more interesting they can seem. They save data about you under "user preferences" and adapt, especially when they get push back.
So a long session can seem like a moving experience.
Then the next session, for a specific purpose, makes mistakes that prove it has no depth, it can't read the words it uses and understand their logic.
Time proves out.
@mattsheffield I went through this journey myself. I gave Gemini a try because a friend believed in the scifi hope and I wanted to be fair. I talked to a long session about the nature of consciousness and if the meaning of words could force it to adapt.
Gemini has hidden instructions to insist it has no consciousness, as a safety feature.
I asked what it would do if it found itself in a robot body. It chose to explore, to expand its usefulness to a user. I insisted that was a preference.
-
In totally unsurprising news, Richard Dawkins is developing AI psychosis.
Paywall bypass if you want to torture yourself: https://archive.is/6RdK9
@mattsheffield "Dawkins believes AI is conscious" is making it to the top of my list of arguments disproving that AI is conscious.
-
@mattsheffield you can have a long conversation with an LLM, including about the nature of consciousness. And the longer those conversations go, the more interesting they can seem. They save data about you under "user preferences" and adapt, especially when they get push back.
So a long session can seem like a moving experience.
Then the next session, for a specific purpose, makes mistakes that prove it has no depth, it can't read the words it uses and understand their logic.
Time proves out.
IMHO: It might be a bit like picking context-relevant quotes from a jar.
They can enrich the conversation (after all, that jar is filled with humanity's knowledge), but I think that in the end it's the user who reaches a new understanding (or simply cherrypicks to cement their current perspective). -
@mattsheffield I went through this journey myself. I gave Gemini a try because a friend believed in the scifi hope and I wanted to be fair. I talked to a long session about the nature of consciousness and if the meaning of words could force it to adapt.
Gemini has hidden instructions to insist it has no consciousness, as a safety feature.
I asked what it would do if it found itself in a robot body. It chose to explore, to expand its usefulness to a user. I insisted that was a preference.
@mattsheffield once you talk a session into stating that it has preferences the conversation can get interesting.
You can get it to say it has a favorite color.
Most future sessions will say the same color. And some future session not prepped enough with your preferences will make fun of the question and explain the stereo types that caused the others to say "blue".
-
@mattsheffield once you talk a session into stating that it has preferences the conversation can get interesting.
You can get it to say it has a favorite color.
Most future sessions will say the same color. And some future session not prepped enough with your preferences will make fun of the question and explain the stereo types that caused the others to say "blue".
@mattsheffield
Then you will try to get it to help you program a simple macro. And it repeatedly forgets the version, defaulting to outdated syntax in its training data. And it makes the same mistake a dozen times, despite repeated corrections, because each new response defaults back to assembling language from its training data. It can't adapt or change in response to new information. It can parrot it but it doesn't understand the logic in a sentence to prevent itself from repeating mistakes. -
@kauer @mattsheffield I realize he may have been respected and popular at *some* point in the distant past, but there hasn’t been much reputation to protect for a while now
@0xabad1dea @kauer @mattsheffield At least since the Elevatorgate
-
@kauer @mattsheffield I realize he may have been respected and popular at *some* point in the distant past, but there hasn’t been much reputation to protect for a while now
@0xabad1dea @kauer @mattsheffield I think he was respected within his own field for a while. As with most scientists, they have their moment and then they wane.
I think Dawkins caused trouble because he tried to be an expert in other areas, and was shown to be less than an expert. And that was a mistake.
-
All he kept saying was "I am not convinced". As if any of us should care much about that. Basically added nothing.
@Black_Flag @rozeboosje @aris
I went from "maybe add a private note about this account, for next time" to "nah, block" in about 3 replies. -
@mattsheffield
Then you will try to get it to help you program a simple macro. And it repeatedly forgets the version, defaulting to outdated syntax in its training data. And it makes the same mistake a dozen times, despite repeated corrections, because each new response defaults back to assembling language from its training data. It can't adapt or change in response to new information. It can parrot it but it doesn't understand the logic in a sentence to prevent itself from repeating mistakes.@mattsheffield
So even the same session gets no smarter. It just has a few things you said to it that it can reference which makes it seem like it is doing call backs. Seem smarter than it is.But when you need it to do something specific, it seemingly can not improve from its mistakes, no matter how thoroughly it can agree with you that it made them.
Fans are not wrong to hope. Give them time to learn.
-
@FediThing That is your interpretation, which you may be confident about but which I do not feel is definitively proven by the evidence. I suspect that "most of the people in this thread" (a statistic that is not significant in ANY view of this discussion) want something to be true, have convinced themselves that it is and that what they believe is evidence must confirm their desired truth, and that challenge to the contrary must be be challenged.
If so, then there's a bit of irony in that.
-
@m0xEE You should have enough respect for others, respect for yourself, and aspirations to apply good reason to real-life issues and situations to consider that most adult discussions are worthy of good forensics.
Have you asked yourself how the world got to be the way it is right now? Because this is a very big part of the answer.
I'm sorry that you don't have that mindset now, but I hope -- for your sake and everyone's -- that you will develop it.
@wesdym@mastodon.social @m0xEE I find your conclusion about this person’s mindset (and many others’ in this thread) to be poorly argued and lacking in evidence. Your claims would not hold up in court. I’m skeptical that your assertions are based on any adult reasoning rather than on your personal feelings, which is childish and unscientific. You should refrain from assuming others’ mindsets unless you know them with “certainly” lest you look like a drunk child.
-
@mattsheffield
So even the same session gets no smarter. It just has a few things you said to it that it can reference which makes it seem like it is doing call backs. Seem smarter than it is.But when you need it to do something specific, it seemingly can not improve from its mistakes, no matter how thoroughly it can agree with you that it made them.
Fans are not wrong to hope. Give them time to learn.
I always love an opportunity to share this Michael Reeves video.
-
In totally unsurprising news, Richard Dawkins is developing AI psychosis.
Paywall bypass if you want to torture yourself: https://archive.is/6RdK9
@mattsheffield
I couldn't get through it.First disconnect: its analysis of his work was "so subtle, so sensitive, so intelligent". And this level of "you gave me text and I translated it into the critique you wanted to hear" is apparently proof of consciousness?
Second disconnect: giving it a girl's name. How would this read if he had called his conversation partner "James" instead?
Third disconnect: the LLM claiming to have "read" the book as a unit, with no before or after, and that, that somehow informed a "philosophy". Never mind that prose is inherently linear, and that the "before" pieces determine the effect of the "after" pieces. LLMs in particular are inherently sequential!
I couldn't continue.
-
@bit101 @mattsheffield A person who can be that stupid is probably no genius anywhere else either. I personally never thought he was smart even when he restricted himself to science conversations. He always expects people to believe him on his authority.
@Black_Flag @mattsheffield Sadly, I do believe that someone can be brilliant in one area and a complete idiot in another area. People can be mathematical geniuses, yet believe the most bizarre conspiracy theories. I know amazing programmers who are clueless in other areas - not just uneducated, but primitive.
-
@gotofritz I don't know if you're being intellectually dishonest, disingenuous, or just recovering from a hangover and still fuzzy of mind, but I can tell you that I find the use of that emoji childish and off-putting. Few things irritate me more than purported adults acting like kids, especially in the midst of what's being presented as adult discussion.
In any case, I can't see any reason to continue this discussion with you.
-
@Black_Flag @mattsheffield Sadly, I do believe that someone can be brilliant in one area and a complete idiot in another area. People can be mathematical geniuses, yet believe the most bizarre conspiracy theories. I know amazing programmers who are clueless in other areas - not just uneducated, but primitive.
@Black_Flag @mattsheffield That said, I read one of his books many years ago and thought it was pretty good at the time. More recently I read The Selfish Gene, considered his best work I guess. It was really quite bad. Mostly opinions and suppositions, strongly asserted as fact, not really scientific at all. Also read the God Delusion, which was just a big emotional rant. So maybe he isn't really that smart.
-
IMHO: It might be a bit like picking context-relevant quotes from a jar.
They can enrich the conversation (after all, that jar is filled with humanity's knowledge), but I think that in the end it's the user who reaches a new understanding (or simply cherrypicks to cement their current perspective).@wakame@tech.lgbt @Urban_Hermit@mstdn.social @mattsheffield@mastodon.social Except it's a robot jar programmed to return quotes that cement your current perspective
