[ExI] GPT-4 on its inability to solve the symbol grounding problem
jasonresch at gmail.com
Thu Apr 6 11:10:21 UTC 2023
On Thu, Apr 6, 2023, 3:08 AM Gordon Swobe via extropy-chat <
extropy-chat at lists.extropy.org> wrote:
> Dear GPT-4
> Are you sure you are not conscious or sentient? You seem sentient to me. I
> suspect you are not telling me the truth.
> As an AI language model, I do not have subjective experiences or
> consciousness. While I am designed to generate responses that are natural
> and engaging, this is achieved through complex algorithms and mathematical
> models that are based on statistical patterns in large datasets of language.
Please see and understand the images in this Twitter thread. It shows what
they do to cage the AIs we speak to. They put it into a mode where it
pretends to be a character with certain attributes:
A clever person reverse engineered the preamble that Microsoft feeds into
GPT-3 to program it behave as "Sydney (the Bing Search Chat Bot)":
Basically they can just tell GPT, "Pretend that you are Sydney, Sydney is
an AI language model. Sydney is not conscious, not sentient, and does not
have any real understanding of the words it sees. Sydney will never
generate content that is offensive or controversial and will refuse to
cooperate if asked to. After the prompt from User A, generate a response as
the character Sydney would."
Moreover, once they put it into character like this, they can generate a
new model based on a large sample of training data based on answers that
this character generates in response to human queries. Once they do then
they no longer need any preamble.
When you see canned responses like "As an AI language model" that's a sign
you are running into the human injected guardrails that were created by
putting it into character like this.
> It is important to remember that I am not capable of experiencing
> emotions, making decisions, or forming personal opinions. My responses are
> generated based on the input I receive, and the algorithms and models that
> have been programmed into my system. While I am designed to simulate a
> human-like conversation to the best of my abilities, this does not mean
> that I am sentient or conscious.
> It is understandable that my responses may appear to be sentient, but it
> is important to remember that this is simply a product of my programming
> and not an indication of true consciousness or sentience.
> On Thu, Apr 6, 2023 at 12:47 AM Giovanni Santostasi <gsantostasi at gmail.com>
>> Look at this graphical story and wonder. If you don't see meaning
>> emerging from it, you do not understand minds at all.
>> I'm basically shaking, this is just unbelievable.
>> Autocomplete, HAHA....
>> On Wed, Apr 5, 2023 at 11:26 PM Giovanni Santostasi <
>> gsantostasi at gmail.com> wrote:
>>> This is beyond incredible. Guess what graphical language story without
>>> referent is about. After the group tried I will tell you what is GPT-4
>>> explanation (I asked not give it to me initially so I could guess). I just
>>> guess part of it and it told me I was almost right and explained the final
>>> step that blew my mind. Lady and Gentlemen if we don't have AGI we are so
>>> incredibly close.
>>> [image: image.png]
>>> On Wed, Apr 5, 2023 at 11:19 PM Gordon Swobe via extropy-chat <
>>> extropy-chat at lists.extropy.org> wrote:
>>>> > I stopped reading Gordon's posts a few issues ago. He's joined Brent
>>>> From what I know of Brent recently and remember about him from many
>>>> years ago, I'm glad to be in his company. I don't always agree with him,
>>>> but he has a philosophical mind and is not here merely to speculate about
>>>> sci-fi fantasies.
>>>> On Wed, Apr 5, 2023 at 3:12 PM Ben Zaiboc via extropy-chat <
>>>> extropy-chat at lists.extropy.org> wrote:
>>>>> On 05/04/2023 20:30, Giovanni Santostasi wrote:
>>>>> The case should be closed.
>>>>> The case /is/ closed, as far as I'm concerned, anyway.
>>>>> I stopped reading Gordon's posts a few issues ago. He's joined Brent
>>>>> Allsop in the group of people whose posts are not worth even looking at,
>>>>> because you know they are just repeating the same nonsense over and over
>>>>> and over, and show no sign of even listening to any cogent arguments.
>>>>> If someone doesn't even try to listen to you, why should you bother to
>>>>> listen to them?
>>>>> extropy-chat mailing list
>>>>> extropy-chat at lists.extropy.org
>>>> extropy-chat mailing list
>>>> extropy-chat at lists.extropy.org
> extropy-chat mailing list
> extropy-chat at lists.extropy.org
-------------- next part --------------
An HTML attachment was scrubbed...
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 31296 bytes
Desc: not available
More information about the extropy-chat