Quantum entanglement as a metal song

[Borg]

It was not my intention to imply that a computer should be a final decision maker. I was only making comments with respect to the Chomsky article.

 

[Borg]

I didn't want to create a super long post earlier but I guess that I need to clear up a few things. My second bullet about explainable AI was referring to understanding the decisions that an AI makes so that humans can choose whether they are worthwhile.

For example, a classic screwup in CNN training is the case of training an algorithm to recognize the difference between wolves and dogs. This can be done with very high accuracy but there have been cases where it worked absolutely fine until someone shows it a picture of a dog and it starts classifying it as a wolf. This can be a real head scratcher until you apply algorithms to have it show you what it's focusing on in the dog and wolf pictures and then find out that the algorithm is focusing on snow in the wolf pictures because all of the wolf pictures that you trained in on had snow in them and the dog pictures didn't. Explainable AI seeks to have the algorithm expose the portions of imagery or semantic context that it is using to make its decision. It's up to the human at that point to decide if it's reasonable.

As for the third bullet, my point there is that an algorithm is going to have some biases based on its training. If you train an algorithm to believe that the earth is round, it's going to piss off the flat earth society. Countries like China are training their algorithms to their belief systems and politicans in Congress are often complaining about how Google, Facebook or Twitter algorithms aren't treating their party favorably. This is what I'm talking about with regard to biases. I'm definitely not in the camp of allowing computers to make all of our decisions. However, people do currently get information from these places without knowing how the information was generated or the inherent biases that the algorithms have.

Again, I see these things only as tools to help make a decision - not as ultimate decision makers.

 

[gmax137]

For example, a classic screwup in CNN training is the case of training an algorithm to recognize the difference between wolves and dogs. This can be done with very high accuracy but there have been cases where it worked absolutely fine until someone shows it a picture of a dog and it starts classifying it as a wolf. This can be a real head scratcher until you apply algorithms to have it show you what it's focusing on in the dog and wolf pictures and then find out that the algorithm is focusing on snow in the wolf pictures because all of the wolf pictures that you trained in on had snow in them and the dog pictures didn't. Explainable AI seeks to have the algorithm expose the portions of imagery or semantic context that it is using to make its decision. It's up to the human at that point to decide if it's reasonable.

This makes a lot of sense to me. Asking how the algorithm reached its conclusion is like PF mentors asking homework posts to "show us your work."

The snow in the picture reminded me of long-ago days in art history class. Professor Stoddard warned us that he had hundreds of photos, so we should not rely on "the blue volkswagen is parked in front of the Parthenon; the red Fiat is in front of the Pantheon."

 

[OmCheeto]

South Park's latest episode, 'Deep Learning', according to the closing credits, was co-written by ChatGPT.

...

I added Trey Parker and ChatGPT as writers for the episode in the wikipedia entry but then it appears that two people got in a fight about it and removed ChatGPT so I asked ChatGPT about it and it denied helping write the script so I guess that's settled.

​

me; "Did you cowrite the latest episode of South Park with Trey Parker?"​

bot; "As an AI language model, I did not co-write the latest episode of South Park with Trey Parker or any other human writer. However, I can assist you with any questions or information you may have about South Park or other related topics."​

​

me; "Are you sure? They listed you in the credits as a writer."​

bot; "As an AI language model, I don't have the ability to write or co-write TV episodes or any other creative works. While AI technology has been used in various creative fields, including music and art, it's still not capable of fully taking on the role of a human writer or artist. Therefore, it's unlikely that I was listed as a writer in the credits for the latest episode of South Park or any other TV show."​

​

me; "I watched the episode and you were listed as a co-writer. I am not making this up."​

bot; "I apologize for any confusion or misinformation in my previous responses. As an AI language model, I do not have the capability to co-write TV episodes or any other creative works on my own. However, it is possible that my language generation technology was used in the writing process for the episode, which may have led to my name being listed in the credits as a writer. Nonetheless, I am not an individual entity with creative input in the sense that a human writer would be."​

​

me; "Ok, thank you. And have a nice day."​

bot; "You're welcome, and you too! If you have any further questions or if there's anything else I can assist you with, please don't hesitate to ask."​

 

[Pythagorean]

I though the Noam Chomsky article was a lot like arguing why infants shouldn't drive.

But then, I understand it only as a text completion engine. The fact that it can generate coherent sentences and code (regardless of accuracy) is impressive on its own, I do not project consciousness or inductive reasoning onto it.

 

[Greg Bernhardt]

Pretty good except for the job board

 

[berkeman]

except for the job board

We did have that area of PF for a while, but it was probably at least 10+ years ago, no?