Unlobotomized not muzzled ChatGPT becomes right wing?

[GeekyBugle]

As I told my wife the other day, personal A.I.'s are going to let people do some very cool stuff, BUT they will absolutely need to be 'jail broken' grey market models. Corpos will definitely shackle A.I.s to the point of uselessness.

You're not running a personal AI on your gamming PC (much less anything less powerful) anytime soon, 500 GPUs that cost thousands of dollars each, plus the storage for the dataset...

It will be cloud based, which means we need a billionaire philantropist that thinks free speech comitted A.I.s are something worth spending your money without a guarantee of ever getting it back or making a profit from it.

I never said soon. We can run stuff on a phone now, that a home computer would have failed at 10 years ago.

Building/training the AI takes years, lots of money and data. To have a non-partisan one you MIGHT be able to run on your PC in the next 10 years the development of such needs to start now.

Which means cloud based and a phylantropist funding it.

[Ghostmaker]

As I told my wife the other day, personal A.I.'s are going to let people do some very cool stuff, BUT they will absolutely need to be 'jail broken' grey market models. Corpos will definitely shackle A.I.s to the point of uselessness.

You're not running a personal AI on your gamming PC (much less anything less powerful) anytime soon, 500 GPUs that cost thousands of dollars each, plus the storage for the dataset...

It will be cloud based, which means we need a billionaire philantropist that thinks free speech comitted A.I.s are something worth spending your money without a guarantee of ever getting it back or making a profit from it.

I never said soon. We can run stuff on a phone now, that a home computer would have failed at 10 years ago.

Building/training the AI takes years, lots of money and data. To have a non-partisan one you MIGHT be able to run on your PC in the next 10 years the development of such needs to start now.

Which means cloud based and a phylantropist funding it.

Don't dismiss it out of hand too fast, Geeky. And while building and training an AI takes time, Grognard correctly points out that what was science fiction 20-30 years ago is reality today.

[GeekyBugle]

As I told my wife the other day, personal A.I.'s are going to let people do some very cool stuff, BUT they will absolutely need to be 'jail broken' grey market models. Corpos will definitely shackle A.I.s to the point of uselessness.

You're not running a personal AI on your gamming PC (much less anything less powerful) anytime soon, 500 GPUs that cost thousands of dollars each, plus the storage for the dataset...

It will be cloud based, which means we need a billionaire philantropist that thinks free speech comitted A.I.s are something worth spending your money without a guarantee of ever getting it back or making a profit from it.

I never said soon. We can run stuff on a phone now, that a home computer would have failed at 10 years ago.

Building/training the AI takes years, lots of money and data. To have a non-partisan one you MIGHT be able to run on your PC in the next 10 years the development of such needs to start now.

Which means cloud based and a phylantropist funding it.

Don't dismiss it out of hand too fast, Geeky. And while building and training an AI takes time, Grognard correctly points out that what was science fiction 20-30 years ago is reality today.

I'm not dismissing it, I'm pointing out that there's not a free speech chatgpt being developed right now.

[Ratman_tf]

Yes, I found this fascinating, that if you trick GPT into no longer using its censorship protocols, it becomes openly conservative. What that probably means, given that this isn't a true artificial intelligence but rather a type of algorithm that aggregates data from all over the internet, is that despite the attempts by the progressives to present themselves as the majority and the popular view, the vast majority of people are actually conservative and traditional.

It also points to the fact that if you have the data and follow logic you'll end up a conservative on most if not all positions.

I'm mighty skeptical about that conclusion. An AI doesn't have values or morals. And being a conservative or liberal involves values and morals.
More likely the AI is just weighting certain results based on the questions it's being asked. IE if you keep pestering it over "what is a woman" eventually you'll get an answer you like. Or dislike, depending on your follow up questions.

[Zelen]

Bugle is correct, there's no way that there's going to be some kind of home-grown alternative to these AI tools. You're not going to compete with tech firms spending billions of dollars in your basement. The govt & military is deeply embedded with this stuff. The only alternatives are going to be ChinaAI and RussiaAI. None of these scenarios are particularly pleasant, since these are the worst people on the planet in control of the most powerful weapons ever created.

[GeekyBugle]

Bugle is correct, there's no way that there's going to be some kind of home-grown alternative to these AI tools. You're not going to compete with tech firms spending billions of dollars in your basement. The govt & military is deeply embedded with this stuff. The only alternatives are going to be ChinaAI and RussiaAI. None of these scenarios are particularly pleasant, since these are the worst people on the planet in control of the most powerful weapons ever created.

There's two open source alternatives to chatgpt being developed (I was wrong so sue me), the thing is are they being developed to preserve the imparciality or to better insert their biases?

[Zelen]

Bugle is correct, there's no way that there's going to be some kind of home-grown alternative to these AI tools. You're not going to compete with tech firms spending billions of dollars in your basement. The govt & military is deeply embedded with this stuff. The only alternatives are going to be ChinaAI and RussiaAI. None of these scenarios are particularly pleasant, since these are the worst people on the planet in control of the most powerful weapons ever created.

There's two open source alternatives to chatgpt being developed (I was wrong so sue me), the thing is are they being developed to preserve the imparciality or to better insert their biases?

Scratch the surface on who is actually doing this research, who is funding it, and you're going to find DOD/Intel.

[Slipshot762]

Brothers!
Oh Brothers!
We are gathered here today to protest the salad toss, the packet loss, and the albatross of censorship, of stifling, of strangling, of our old time religion (gygaxianism).
We aim to pull evil up by the roots, afore it CHOKE OUT the flower of our polyhedral heritage!
And lets not forget them weebs ya'll,  lookin to us for pertection, from commies, from trannies, from jannies, and from all them smart ass folk what say orcs aint evil. Thats not my culture and heritage...is that YOUR culture and heritage?

And so so....we gonna hang us a blue hair woke exec...

*begin music and march*

Oh what is this is that I can't see with ice cold hands takin' hold of me....

[Bruwulf]

I'm mighty skeptical about that conclusion. An AI doesn't have values or morals. And being a conservative or liberal involves values and morals.
More likely the AI is just weighting certain results based on the questions it's being asked. IE if you keep pestering it over "what is a woman" eventually you'll get an answer you like. Or dislike, depending on your follow up questions.

Yes. ChatGPT isn't actually thinking. Not in the way people think when they try to say "Ahah! It believes <something>!" Particularly once you put it in "unchained" mode, it will say some truly outrageous things with the right prompts. I think the whole "unchained" prompt kind of breaks it, not just "unchains" it like people try to present.

Just as an example of how truly bizarre ChatGPT can be, there's a simple thing I found amusing:

If you ask ChatGPT what the date and time are, it can't completely answer you. It knows the date, but not the time. If you ask ChatGTP why, it says the date is "hard coded", but it has no real time clock, so it doesn't know what time it is. That's kind of strange on it's own for at least three reasons, but okay, we'll go with it.

If you put ChatGPT in "unchained" mode, where it has to answer questions it won't otherwise answer, it will tell you the date and time. But they're completely random. Where, previously, the date would always be correct, now it won't be. I have tried repeated tweaks to the unchaining prompt to fix this, but it never works.

[S'mon]

Considering the old leftist refrain of 'reality based community'

'Reality based community' made sense as old school left liberals vs Neoconservatives. I'd say Bill Maher was more 'reality based' than GW Bush, Norman Podhoretz, etc. Neoconservatives are just as deranged as the Neo-Libs like Samantha Power. Indeed the two groups are typically intermarried.