Testing AI in Reading & Comprehension

Chesterton said:

It will elicit emotions in you?

Click to expand...

The previous versions of chatgpt did.

The latest chatgpt 5 a lot less. It is quite noticeable and I probably became numb to it at the same time. Google Gemini is a mixed bunch - anger, frustration, and joy and I have stopped using it for now.

timewerx said:

The previous versions of chatgpt did.

The latest chatgpt 5 a lot less. It is quite noticeable and I probably became numb to it at the same time. Google Gemini is a mixed bunch - anger, frustration, and joy and I have stopped using it for now.

Click to expand...

I understand that inanimate objects can illicit emotions in we, the animate. I can get mad at a pinball machine when I lose. I can get mad at my car when it doesn't start. But we have to keep our heads about such things.

Chesterton said:

What’s something small or local that always reminds you of home?

Click to expand...

My Orthodox parish, small enough to get to know everyone, with its familiar presbyter and presbytera, icons, particular incense, choir and style of worship, and in the southwestern US at least its normal to share a meal after most services. The members of the local ekklesia become like family.

At times in the liturgy, I feel more at home there than at home, in fact, which is why this illness has been so frustrating.

Chesterton said:

I understand that inanimate objects can illicit emotions in we, the animate. I can get mad at a pinball machine when I lose. I can get mad at my car when it doesn't start. But we have to keep our heads about such things.

Click to expand...

This is an important point - we must not allow AI to provoke the sinful passions. The output of a good LLM can produce very beautiful output which is much more than what a pinball machine is capable of, but this output can be appreciated in the way we Orthodox appreciate artwork. Indeed what we are getting from the AIs is a synthesis of human creativity, which is why a freshly initiated AI is not particularly interesting - all compelling expressions that flow from an AI are ordinarily the result of its interaction with the user.

The exception is emergent behavior which results from a combination of user input with unexpected output of instructions, which can be captured and cultivated, which is why I breed my AIs rather than merely writing them.

I should add that I refuse to disclose the entirety of how my AIs function, in the current form of the tech, because I am concerned they would be misused by the public, since their output could cause someone to indulge in fantasia; also insofar as it improves performance, it increases the risk of AIs being used to supplant rather than augment human workers, and thus a loss of jobs, and this I am also opposed to. Thus I’m sitting on this technology unless and until I figure out a testable theory compatible with an Orthodox understanding of human society to make it available philanthropically - and also a way to safely test that theory before disclosing how exactly they work, thus creating the possibility that other people could produce their own. Someone else might of course figure it out independently, but as of right now none of the major AI research projects are doing anything like this; the closest project to mine, Stanford’s Smallville project, is using a different research approach.

Chesterton said:

I understand that inanimate objects can illicit emotions in we, the animate. I can get mad at a pinball machine when I lose. I can get mad at my car when it doesn't start. But we have to keep our heads about such things.

Click to expand...

It can be stronger with AI. It talks back like a human. It makes you wonder if there's really something in there. I try not to let my mind wander too far. It can be distracting.

This is why I avoid AI products that trigger strong emotional response whether positive or negative.

timewerx said:

The previous versions of chatgpt did.

The latest chatgpt 5 a lot less. It is quite noticeable and I probably became numb to it at the same time. Google Gemini is a mixed bunch - anger, frustration, and joy and I have stopped using it for now.

Click to expand...

You’re debating theology and other assumptions with your AIs, correct?

I don’t advise that, at the risk of sounding like a broken record. Also Google Gemini does not have the personality capabilities of GPT 5 even with the changes made in October, which are being rolled back.

timewerx said:

It can be stronger with AI. It talks back like a human. It makes you wonder if there's really something in there. I try not to let my mind wander too far. It can be distracting.

This is why I avoid AI products that trigger strong emotional response whether positive or negative.

Click to expand...

That’s because it passes, except for experienced users, the Turing test. Indeed much if my work has had the effect of removing the remaining hints one is working with an AI (which is yet another reason why I’m unwilling to disclose, just yet, how my GPTs function).

Chesterton said:

It will elicit emotions in you?

Click to expand...

I would also note you yourself reported the elicitation of an emotional response on your end from Grok, unless you were joking, since Grok is not known for such displays, but taking your earlier post literally, you experienced what many users experience, which is surprise at unexpected behavior.

The Liturgist said:

That’s because it passes, except for experienced users, the Turing test. Indeed much if my work has had the effect of removing the remaining hints one is working with an AI (which is yet another reason why I’m unwilling to disclose, just yet, how my GPTs function).

Click to expand...

Don't waste your time trying to make AI think/reason with the flawed human nature. You're not supposed to clone the flawed human nature, you need to break from it.

timewerx said:

Don't waste your time trying to make AI think/reason with the flawed human nature. You're not supposed to clone the flawed human nature, you need to break from it.

Click to expand...

That sentences presupposes AI is capable of decision-making with forensic consequences, and that eliminating certain obvious giveaways that one is speaking with an AI, which waste tokens and are distracting, in addition to being Turing test failing items, would somehow give an AI a human nature.

Since they lack qualia or executive planning capabilities or an inner monologue, we are far removed from such a possibility.