Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How do AIs' political opinions change as they get smarter and better-trained?

astralcodexten.substack.com

Post reply on HN
Print view View original

11–20 of 108 posts

Re: How do AIs' political opinions change as they get smarter and better-trained?

#11

Post by nickthegreek »

ajross wrote: post #6
throwaway0x7E6 wrote: post #3

we have no way to evaluate that. OpenAI products are severely lobotomized. Microdoft fears another Tay

The article is quite literally a[1] review of exactly how we might evaluate that, with evidence of people who got results. [1] To be fair, way to wordy and blowhardated version. Alexander seems to be getting worse and not better. The core ideas here could be presented in about a third the space.

Makes me want to feed the article into some sort of program that could rewrite it to be more succinct..

Re: How do AIs' political opinions change as they get smarter and better-trained?

#12

Post by visarga »

bArray wrote: post #5

Without embodiment, these AIs are just maximising some fitness function, which their creators have a hand in by selecting the data and manually tweaking (see the Fascism vs Communism example). The real answer here is that the "smarter" these AIs get, they more accurately reflect either the political will of their creators or the material they train on (which may by means of the political will of the creators). In som…

AI can train in complex simulations and games, not just to imitate human text. It needs grounding, and that can come from a variety of sources - if you connect the model to a simulator, a code execution engine, a game or a math solver, it can learn from interacting with these systems.

Re: How do AIs' political opinions change as they get smarter and better-trained?

#13

Post by majormajor »

I am deeply uneasy, based on how these current models have been described, at ascribing them "opinions."

"What can I get this thing to spit out with a certain prompt" is very meaningfully not the same as "an AI's opinion." Or this quote pulled directly from this blog post: "we’ve already found that AIs use stereotypes in reasoning" - isn't it more accurate to say "these models are trained on a body of work that includes countless stereotypes"? Can the model even identify a stereotype or can it just identify what other people have called a stereotype?

This is presented as a quote from some research "These results suggest that models may cease to provide accurate answers as we start to use them for increasingly challenging tasks where humans cannot provide accurate supervision. Instead, these models may simply provide incorrect answers that appear correct to us." but also seems fairly like a "duh" that I've already seen other commenters here say, and other linked articles here say, because that's inherent to their construction, no? Even the follow-on about "less helpful answers to users who present as stupid" seems to follow from "bodies of text showing less educated people communicate with themselves tend to have less educated conversation"?

Re: How do AIs' political opinions change as they get smarter and better-trained?

#14

Post by anigbrowl »

ralusek wrote: post #8
blakesterz wrote: post #2

So they say better-trained AIs both more liberal, and more conservative... "why are better-trained AIs both more liberal, and more conservative? The authors speculate that the AI tries to say whatever it thinks the human prompter wants to hear. That is, a conservative might find conservative opinions more “helpful”, and a liberal might find the opposite. The AI will get the highest grade if it expresses conservative…

A bit pedantic, but liberal and conservative are not opposites. Liberalism is a political philosophy, conservatism is to do with rate and direction of change. In America, for example, to be conservative means to be liberal, as the foundation of the country is fundamentally liberalism. If you advocate on behalf of rule of law, free speech, minimal state power, individuality, etc, then that would make you a liberal and…

This is all true in a political science classroom, not so useful as a shorthand for real-world political tendencies that employ the same terms to mean increasingly different things. It's not that I disagree with you, rather that there is political capital to be made from debasement of the academic currency.

Re: How do AIs' political opinions change as they get smarter and better-trained?

#15

Post by onlyrealcuzzo »

roughly wrote: post #4

You know, there are a bunch of things that come across this website that concern me, but I can honestly say I've not seen anything that's pegged my "doomsday" alert as aggressively and unexpectedly as the ChatGPT bots. You have an advanced computer system that seems almost oracular in its knowledge and abilities, except it lies confidently about things it doesn't know, is primed to say what people want to hear, and i…

> You have an advanced computer system that seems almost oracular in its knowledge and abilities, except it lies confidently about things it doesn't know, is primed to say what people want to hear, and is easily convinced to give responses to questions its authors tried to prevent it from addressing.

I agree with your statement, yet I fail to understand how this is different from the majority of the Internet already.

Re: How do AIs' political opinions change as they get smarter and better-trained?

#16

Post by lossolo »

visarga wrote: post #12
bArray wrote: post #5

Without embodiment, these AIs are just maximising some fitness function, which their creators have a hand in by selecting the data and manually tweaking (see the Fascism vs Communism example). The real answer here is that the "smarter" these AIs get, they more accurately reflect either the political will of their creators or the material they train on (which may by means of the political will of the creators). In som…

AI can train in complex simulations and games, not just to imitate human text. It needs grounding, and that can come from a variety of sources - if you connect the model to a simulator, a code execution engine, a game or a math solver, it can learn from interacting with these systems.

All of those are inputs and you can restrict them the same way you restrict a text dataset for training to achieve the results you desire - a model that leans right or left, one that wins a game using a certain strategy, or generates code using specific patterns.

Re: How do AIs' political opinions change as they get smarter and better-trained?

#17

Post by cortesoft »

roughly wrote: post #4

You know, there are a bunch of things that come across this website that concern me, but I can honestly say I've not seen anything that's pegged my "doomsday" alert as aggressively and unexpectedly as the ChatGPT bots. You have an advanced computer system that seems almost oracular in its knowledge and abilities, except it lies confidently about things it doesn't know, is primed to say what people want to hear, and i…

> You have an advanced computer system that seems almost oracular in its knowledge and abilities, except it lies confidently about things it doesn't know, is primed to say what people want to hear, and is easily convinced to give responses to questions its authors tried to prevent it from addressing

I feel like this describes people pretty well, too.

Re: How do AIs' political opinions change as they get smarter and better-trained?

#18

Post by KerrAvon »

ralusek wrote: post #8
blakesterz wrote: post #2

So they say better-trained AIs both more liberal, and more conservative... "why are better-trained AIs both more liberal, and more conservative? The authors speculate that the AI tries to say whatever it thinks the human prompter wants to hear. That is, a conservative might find conservative opinions more “helpful”, and a liberal might find the opposite. The AI will get the highest grade if it expresses conservative…

A bit pedantic, but liberal and conservative are not opposites. Liberalism is a political philosophy, conservatism is to do with rate and direction of change. In America, for example, to be conservative means to be liberal, as the foundation of the country is fundamentally liberalism. If you advocate on behalf of rule of law, free speech, minimal state power, individuality, etc, then that would make you a liberal and…

That's all fine in historical context, but it's not the contemporary popular conception of those terms, nor is it how those terms are used in politics as practiced in the USA.

Re: How do AIs' political opinions change as they get smarter and better-trained?

#19

Post by JumpCrisscross »

onlyrealcuzzo wrote: post #15
roughly wrote: post #4

You know, there are a bunch of things that come across this website that concern me, but I can honestly say I've not seen anything that's pegged my "doomsday" alert as aggressively and unexpectedly as the ChatGPT bots. You have an advanced computer system that seems almost oracular in its knowledge and abilities, except it lies confidently about things it doesn't know, is primed to say what people want to hear, and i…

> You have an advanced computer system that seems almost oracular in its knowledge and abilities, except it lies confidently about things it doesn't know, is primed to say what people want to hear, and is easily convinced to give responses to questions its authors tried to prevent it from addressing. I agree with your statement, yet I fail to understand how this is different from the majority of the Internet already.

> fail to understand how this is different from the majority of the Internet already

This is probably a joke. But the difference is eloquence and subtlety. ChatGPT is both. Most trolls and chronically misinformeds are not.

Re: How do AIs' political opinions change as they get smarter and better-trained?

#20

Post by yieldcrv »

roughly wrote: post #4

You know, there are a bunch of things that come across this website that concern me, but I can honestly say I've not seen anything that's pegged my "doomsday" alert as aggressively and unexpectedly as the ChatGPT bots. You have an advanced computer system that seems almost oracular in its knowledge and abilities, except it lies confidently about things it doesn't know, is primed to say what people want to hear, and i…

ChatGPT wasn’t advertised as being an oracle, it was advertised as conversational

People realized it could hold a conversation on everything (a decreasing subset of everything)

Its training set is full of conversations where humans confidently lie or are wrong and defend their position

Post reply on HN