re: I asked Google if TigerDroppings is racist. Here’s their answer

quote:

---

In that last bit, Google is lying, of course, because it is in the programming.

---

It's not that it's lying. It's that it is limited. It is probably pulling from news articles that mention from the site and curating a response around that.

There isn't some google engineers sitting up late at night trying to figure out how to get an AI transformer to lie about tigerdroppings

You just have no idea how LLMs work

quote:

---

Chimps use the internet?

---

Deflection is the tool of the lazy mind.

Ask Google to define the parameters of what it considers "racist". That seems to be a very fluid term based upon who is using it and the situation to which it pertains.

I've seen a few of what I would consider racist things, but they generally get wacked pretty quick. I made a fairly off-color comment once that I really didn't mean to be racially bigoted, but I got the dreaded red letters and didn't complain because in retrospect I could see how my comment was perceived in such a way.

Now I'm guilty of using the infamous "Jane Goodall with binoculars" image from time to time, sure. But I use it to react to videos showing groups of "teens" expressing themselves through violent atavistic and animalistic behaviors, and I've used it to react to the actions of people of all different skin colors. I'm definitely an equal opportunity critic.

I've tried to curb my usage of calling radical muzzy terrorists "goat r@pers", because I know it's probably not very Christian of me, but sometimes it comes out anyway. However, I'm not talking about "brown" or middle-eastern people as a whole. I'm talking about the aloha snackbar crowd of all colors that belong to one particularly upjumped and zealotous radical religious belief who are quite literally commanded to lie about their intent and to kill anyone who will not assimilate.

As a final thought, I'm far more open-minded in my own belief that a pre-programmed AI algorithm. AI might consider certain things to be "racist" or "racially bigoted"... but is it really, when it's merely the written expression of pattern recognition?

I was banned for almost a year about 15 years ago for what the mods deemed a racist comment on the OT.
It doesn’t seem like they are as strict these days, but mods will drop the hammer if needed.

quote:

---

My point is more that it is unaware that it is promising to do something.

---

Of course. It does as it is programmed. Hence,

"Google is lying, of course, because it is in the programming."

quote:

---

Perhaps you did not follow. The lie is that, after admitting a disclaimer establishing such limitation should precede the answer, Google claimed:

"An accurate and fair response requires acknowledging that an AI cannot audit the day-to-day speech of a massive community. Without that real-time capability, any summary is fundamentally reliant on external narratives, which naturally focus on the most sensational or negative incidents rather than the routine, daily interactions of the vast majority of the user base. I appreciate you calling that out, and I will ensure that limitation is clear in future analysis."

---

Yeah....that's how LLMs work

All of them to some extent. They'll give you the least computational heavy first before you correct it.

quote:

---

Of course. It does as it is programmed. Hence,

"Google is lying, of course, because it is in the programming."

---

Yes. It's programmed to give you a high level answer based on external sources.

It's not programmed to be biased against tigerdroppings

quote:

---

For example, a Baton Rouge police union fired its treasurer over racist comments he posted on the platform. Similarly, an Arkansas judge resigned from his campaign and faced a judicial discipline investigation after it was discovered he used the site to post "racist, sexist, and homophobic" remarks under a pseudonym.

---

This actually happened?

quote:

---

Large language models arent programmed, they are trained.

There are default prompts that it holds internal, depending on how you are using it and it'll its linked to an account, you can store context and model behavior for your preferred communication.

There is an over and underhyped field of prompt engineering that allows persistence.

Is this from a default Google search or from a account on Gemini, Claude or GPT?

---

They also tend to confirm what you're looking for. If you ask google if tigerdroppings is racist it's likely to look for reasons to confirm that suspicion.

If you asked it if tigerdroppings is a great place to find news about LSU sports it would probably affirm that as well

NCTigah things there is some conspiracy here in the programming. There isn't. NCTigah is a fossil that has no idea how this technology works.