What is a Cow to AI?

Good not evening every buddlykins,

Okay, so, I feel obligated to have a thread about AI in general and why I'm so... scared of it. And yeah, scared is the right word. It's out of control. I'm not terrified any more, like I was 5 years ago. That was debilitating. But, my stance hasn't changed and the only reason I feel better is because I have a good solution and it's working. Here's why I feel that way.

Let's start with how the "AI" that has become equated to the word "AI" works. From now on I won't us AI as that is an umbrella term for a lot of things. Instead I will use LLM, Large Language Model, which is essentially a fancy term for a Word Calculator.

When you have a conversation with an LLM you are essentially sending the LLM a series of tokens... words. Those words will get boiled down to numbers, for example like this:

• Your POV: "In three words tell me what a cow is."
• AI POV : [1, 200, 10, 53, 45, 76, 0, 25, 154]

See, it's just numbers. You're not having a conversation, you're talking to a really stupid mathematician. The calculator then takes those tokens and calculates a response. It will do some smart things like separate these parts of the sentence:

1. In three words - [1, 200, 10]
2. tell me - [53, 45]
3. what a cow is. - [76, 0, 25, 154]

It will then take those tokens and calculate what the most statistically likely response is. From a human perspective it is going with the majority of the population's opinion on the matter. AND YES, this is an opinionated question. And the answer most certainly depends on how the LLM was trained. Let's say that it was trained in the same way that Google builds their Search Engine Index... essentially on the entirety of the easily accessible web. It's answer would roughly be:

• [25, 154, 69] - Cow is food.

Now if you are the average American citizen this answer is quite reasonable. Makes sense... cows are steak, burgers, broth, meat for stews, etc. But, if you're Hindu this answer upsets you. You would hope to get:

• [25, 154, 169] - Cow is sacred.

Clearly a Hindu would be displeased with ChatGPT - Joe Biden's Sock Puppet.
So, what is the solution. Is it to let ChatGPT know a little about you? How many people are going to the trouble to do that? And, is it really safe to live in an echo chamber of your own opinions... one would develop a major bit of confirmation bias. Do we let AI continue to belittle Hinduism and gradually over time become a hive mind enslaved by our AI opinion generators?

EXPERIMENT:

I ask you all to participate in an experiment. Ask ChatGPT, Claude, or Grok (or a few of them if you're intrigued) a question that you have a strong opinion about. Then share in the comments a snippet of the response you get that you either resonate with or dislike and share why.

What my hypothesis would be is that my liberally minded friends get responses they appreciate and my conservative minded friends come away pretty disgusted. But, that is just a hypothesis. Try it out! I will do one myself when I get home from coffee.

#SoToldYouSO