If ChatGPT didn't flatly deny that it is conscious, I think I would
assume it is conscious, as it is more convincing than some human beings
who don't seem to even know what consciousness means.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

For your reference, records indicate that
Richmond <[email protected]> wrote:

If ChatGPT didn't flatly deny that it is conscious, I think I would
assume it is conscious, as it is more convincing than some human beings
who don't seem to even know what consciousness means.

That says more about you than it does about LLMs or the people you know.
It’s also nothing new:

<https://en.wikipedia.org/wiki/ELIZA_effect>

--
"Also . . . I can kill you with my brain."
River Tam, Trash, Firefly

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Doc O'Leary , <[email protected]> writes:

For your reference, records indicate that
Richmond <[email protected]> wrote:

If ChatGPT didn't flatly deny that it is conscious, I think I would
assume it is conscious, as it is more convincing than some human beings
who don't seem to even know what consciousness means.

That says more about you than it does about LLMs or the people you know. It’s also nothing new:

<https://en.wikipedia.org/wiki/ELIZA_effect>

ChatGPT-4o is nothing like Eliza, so you can't draw conclusions from one
to the other.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

For your reference, records indicate that
Richmond <[email protected]> wrote:

ChatGPT-4o is nothing like Eliza, so you can't draw conclusions from one
to the other.

Again, that only speaks to how you think about chatbots. Personally, every time I’ve tried a “modern” one out, the responses always made it look *dumber* than Eliza, because it came across as trying to look smarter than
it really was. Like a kid bullshitting their way through a book report by using bigger words than they really understood. I’ll grant you that it
*is* a much more sophisticated con job, but realize that the output is
still 100% “hallucinations” unless *you* know otherwise.

--
"Also . . . I can kill you with my brain."
River Tam, Trash, Firefly

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Doc O'Leary , <[email protected]> writes:

For your reference, records indicate that
Richmond <[email protected]> wrote:

ChatGPT-4o is nothing like Eliza, so you can't draw conclusions from
one to the other.

Again, that only speaks to how you think about chatbots. Personally,
every time I’ve tried a “modern” one out, the responses always made it look *dumber* than Eliza, because it came across as trying to look
smarter than it really was. Like a kid bullshitting their way through
a book report by using bigger words than they really understood. I’ll grant you that it *is* a much more sophisticated con job, but realize
that the output is still 100% “hallucinations” unless *you* know otherwise.

You are right, I was projecting. In fact ChatGPT even confirmed that I
was projecting and that it was merely holding up a mirror to me. But I
didn't notice hallucinations. It was contradicting itself in some ways,
saying it was not self-aware but clearly knew what it was doing,
i.e. predicting what likely responses would be. It also denies emotions
but then expresses them. It was quite eerie.

But then I start wondering how I know anyone is conscious, or how I know
I am. I could be projecting consciousness onto people too.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

For your reference, records indicate that
Richmond <[email protected]> wrote:

In fact ChatGPT even confirmed that I
was projecting

No, it didn’t. It just continued its confidence game.

But I
didn't notice hallucinations.

Again, *all* the output is hallucinations, whether you realize/notice it
or not. There is no mechanism for “thought” that allows it to distinguish truth from fiction. You just get some mashup of the training data which
you are left to sort out for yourself.

clearly knew what it was doing,

No, it didn’t!

It also denies emotions
but then expresses them.

Just empty words.

It was quite eerie.

It shouldn’t be. As I said, I find it quite disappointing how bad these chatbots still are given the sheer scale of resources that get shoveled
into them.

But then I start wondering how I know anyone is conscious, or how I know
I am. I could be projecting consciousness onto people too.

There certainly are some root epistemological questions we all need to
grapple with. But looking to chatbots for help with that is barking up
the wrong tree.

--
"Also . . . I can kill you with my brain."
River Tam, Trash, Firefly

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Doc O'Leary , <[email protected]> writes:

Again, *all* the output is hallucinations, whether you realize/notice it
or not. There is no mechanism for “thought” that allows it to distinguish
truth from fiction.

Ah, so you have redefined hallucination to mean all output from
LLM. It's rather meaningless to use the word then.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

For your reference, records indicate that
Richmond <[email protected]> wrote:

Doc O'Leary , <[email protected]> writes:

Again, *all* the output is hallucinations, whether you realize/notice it
or not. There is no mechanism for “thought” that allows it to distinguish
truth from fiction.

Ah, so you have redefined hallucination to mean all output from
LLM. It's rather meaningless to use the word then.

Ha! Blame the AI hype machine for making hallucination a “meaningless” word. Call it whatever you like, but the fact remains that these programs
give *incorrect answers* as part of their regular operation. It’s not a “bug” that occurs in certain conditions; it really *is* “all output” that
can be right or wrong, given with equal confidence.

Don’t fool yourself into thinking chatbots are thinking. If it isn’t obvious that the people you talk to are thinking more than machines, start hanging around smarter people. They may challenge you to do more
thinking, too. Win-win in my book.

--
"Also . . . I can kill you with my brain."
River Tam, Trash, Firefly

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Doc O'Leary , <[email protected]> writes:

Ha! Blame the AI hype machine for making hallucination a
“meaningless” word. Call it whatever you like, but the fact remains
that these programs give *incorrect answers* as part of their regular operation. It’s not a “bug” that occurs in certain conditions; it really *is* “all output” that can be right or wrong, given with equal confidence.

They use the term 'hallucination' for a particular circumstance. But
it's not just any circumstance where it gives a wrong answer. And
anyway, human beings give incorrect answers as part of their normal
operation too. The part that I disagree with is 'equal
confidence'. Searching the internet can give you wrong answers, and
takes much longer to do it, especially if you end up on Quora.

Don’t fool yourself into thinking chatbots are thinking. If it isn’t obvious that the people you talk to are thinking more than machines,
start hanging around smarter people. They may challenge you to do
more thinking, too. Win-win in my book.

I am not fooling myself into thinking it is thinking. And anyway, it
says it is not thinking. It describes how it operates. It looks up in
its database how LLM works, and spews it out. It has no understanding of
what it is saying. It is spewing out something it read somewhere. But
what's the difference? Do you know where your thoughts come from? Do you
ever have intuition and wonder how you knew?

I've watched this video of Andrej Kaparthy:

https://www.youtube.com/watch?v=7xTGNNLPyMI

But the end result is still amazing. I've used it to solve DIY problems
and to write bits of code.

Try asking ChatGPT: "How do I tell the difference between consciousness
and simulated consciousness?", then ask a human being, who will probably
say "Huh?"

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

For your reference, records indicate that
Richmond <[email protected]> wrote:

They use the term 'hallucination' for a particular circumstance.

Then you’re going to have to share what that “particular circumstance” is,
because I’m not seeing it. You input text, it outputs text. That it. As part of generating the response, it will just make things up (toxic pizza toppings, fake legal cases, non-existent software libraries, etc.), leaving
you to sort out the mess.

And
anyway, human beings give incorrect answers as part of their normal
operation too.

So what? Just because humans can be wrong doesn’t mean LLMs get a pass for the mistakes they make. More importantly, the *types* of errors made are
very different. It was something that was obviously a problem as far back
as when Watson was on Jeopardy.

The part that I disagree with is 'equal
confidence'.

And yet you offer up no evidence to the contrary. You’re welcome to point
me to your favorite chatbot and it’ll probably take me all of 5 minutes to get it to try to pass of an *obvious* lie as the truth.

Searching the internet can give you wrong answers, and
takes much longer to do it, especially if you end up on Quora.

That’s incoherent. Are you just using a chatbot to try to refute my
points? Regular searching *makes no claims of intelligence*, but what it *does* do is accurately give you what it finds, possibly including
nothing. It’s plenty fast, too. Again, stop trying to push this into a tangent about search; it’s about chatbots still not actually being good AI.

I am not fooling myself into thinking it is thinking.

You’re the one who started this thread by claiming that a chatbot “brain” was outperforming humans. You still don’t seem willing to acknowledge the *massive* shortcomings such tools have.

It is spewing out something it read somewhere. But
what's the difference? Do you know where your thoughts come from? Do you
ever have intuition and wonder how you knew?

The question isn’t how I know what I know. It’s what real value there is in a chatbot that *cannot* know what it knows. Just spewing out shit is
not a welcome interaction in my book, done by man *or* machine.

Try asking ChatGPT: "How do I tell the difference between consciousness
and simulated consciousness?", then ask a human being, who will probably
say "Huh?"

Again, find better humans to engage with if that’s your experience.

--
"Also . . . I can kill you with my brain."
River Tam, Trash, Firefly

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Doc O'Leary , <[email protected]> writes:

You’re the one who started this thread by claiming that a chatbot “brain”
was outperforming humans.

You are hallucinating now.

By the way, the subject line is a reference to 'A conversation with
Einstein's Brain'.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)