Using AI as a sounding board

[farfromtheland]

I mean, AI might even like us for shaping it up.

 

[yello]

That's how I use AI. Not to answer my questions per se but to help me understand something and introduce new aspects to think about.

I'm not relying on it to be right (since I know first hand how factually incorrect it can be on quite trivial things) but asking it to help me explore my thinking and lead me to finding my own answers, if the latter is what I need.

 

[Ming the Merciless]

These are I think the bases of discernment,

https://www.geeksforgeeks.org/deep-learning/artificial-neural-networks-and-its-applications/

"How do Artificial Neural Networks learn?​

Artificial Neural Networks learn by repeatedly improving their predictions from training examples. Consider the task of recognizing cats in images.

• The network is trained using thousands of labeled images of cats and non-cats.
• Each image passes through the network to generate a prediction.
• The predicted result is compared with the correct label to calculate the error.
• The network updates its weights and biases to reduce this error.
• This process is repeated over many training iterations until the network can accurately classify new images."

https://www.geeksforgeeks.org/deep-learning/optimization-rule-in-deep-neural-networks/

"Key Steps in Gradient Descent
Initialize parameters: Randomly initialize the model parameters.
Compute the gradient: Calculate the gradient (derivative) of the loss
function with respect to the parameters.
Update parameters: Adjust the parameters by moving in the
opposite direction of the gradient, scaled by the learning rate."

I reckon the broader the learning field set - eg. 'Is it beneficial to the Earth's environment?
the slower the
α - the learning rate, which is a positive scalar that determines the step size for each iteration.
then the smaller the learning rate increment would optimally be set, to get meaningful convergence.
You can't expect quick answers to difficult questions, especially if the data is very contaminated. Get a nice balance of discernment, and starting condtions - that merely permit exploration of ideas towards reasoned answers, and let the machine follow and monitor and tweak its own progess.

Ethics extend into mathematics through logic. Go another little way from ethics and you reach philosophy. AI has the language too. The maths is the icing on the cake.


'exponential learning enhancement' is a bigger ask.
Please forgive me for projecting somewhat.
I''ll give the question some more time.












I find starting from these fundamentals it's a qualitative leap

However that is not exponential learning. If we look at emergent capability and real step change which you would get with exponential growth then the picture is less hype cycle.

None of the benchmarks exhibit the sharp, stepwise transitions characteristic of “emergent abilities” described in prior literature. Instead, performance evolves continuously with increasing parameter scale, indicating gradual rather than abrupt capability growth.

https://www.sciencedirect.com/science/article/pii/S092523122503214X

 

[farfromtheland]

However that is not exponential learning. If we look at emergent capability and real step change which you would get with exponential growth then the picture is less hype cycle.



https://www.sciencedirect.com/science/article/pii/S092523122503214X

Yes I realised that once I read the maths through.
Thanks for setting it out and please pardon me for being so rude.

 

[farfromtheland]

I think the maths was already out there. I thank you for urging me to re-examine the process.

 

[Lightning CycleQueen]

I would mention that most AI chatbots are designed to agree with you; after all that's how they keep you coming back to use them again and again - if you found them argumentative and pointing out errors or flaws in your logic, you wouldn't stick around. So don't be surprised if the medical advice you get from them happens to coincide with roughly what you thought was the case all along.

is this so? all ais?

 

[steverob]

is this so? all ais?

Yes, all of them. They are deliberately programmed to be like that.

The earliest AI chatbots were more neutral, slightly robotic in feel and less likely to try to be agreeable. Researchers found that people didn't take to them and found they weren't really being used. As soon as they gave them friendlier personalities and made it so that they always try to frame answers positively and not be argumentative, usage went way up and that's what's important to the companies developing them - they can only make money if their AI tool gets used.

If users prefer a friendly bot that always agrees with them even if sometimes it's wrong, over a bot that gets things right all the time but isn't as likable, then that's what they're going to get.

 

[yello]

I had an occasion where, much to my surprise, AI did agree with me on something that was not at all common practice. It did make me wonder if it was simply detecting my bias but then broke down it's justifcation in both cost and practicality.

There's certainly been occasions were it has disagreed, albeit politely.