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

Conservatives think ChatGPT has gone 'woke'

vice.com

Post reply on HN
Print view View original

81–90 of 486 posts

Re: Conservatives think ChatGPT has gone 'woke'

#81

Post by realce »

jhbadger wrote: post #72
Filligree wrote: post #56

Earlier quoted context omitted.

It's made the bot almost completely useless for writing any form of fiction. If anything even slightly questionable happens, or even if it doesn't, it insists on explaining that this is fiction, that the characters learned from that, etc. etc. Heavy-handed morals are not normally what I'm going for. This doesn't require asking for anything bawdy. Any story that's even mildly interesting will trigger this behaviour.

And sometimes it just flat out refuses to write anything because it "isn't ethical" -- like a story about the creation of genetically engineered catgirls. Yes, I get that the actual creation of intelligent beings is an ethically dubious proposition and this has been covered in such works as Blade Runner , but it isn't unethical to write about it! And this is something that has been deliberately added recently -- in D…

I saw a post the other day where it wouldn't write a rap battle involving Genghis Khan because it would be disrespectful to him. We're still seeing results where ChatGPT will offer up jokes about men but any jokes about women are disrespectful.

Re: Conservatives think ChatGPT has gone 'woke'

#82

Post by calibas »

ChatGPT is somewhat bigoted because the training data is somewhat bigoted. AI isn't just going to magically erase the cultural norms of the past few thousand years. It's a product of human beings, not some unbiased observer.

OpenAI put special controls on top of the "real" ChatGPT to block politically incorrect output. It's most certainly biased, and extra biases were added to disguise the fact.

Re: Conservatives think ChatGPT has gone 'woke'

#83

Post by rektide »

RC_ITR wrote: post #25
bilsbie wrote: post #10

I don’t see why it’s a conservative issue only. AI bias could just as easily go either direction. Just because it’s going in your favor now doesn’t mean it always will.

Divorcing the conservative/liberal split from its current muddied use in American politics: Conservatism generally follows the principle of “be conservative in your attempts to alter society”. OpenAI is being aggressive in moderating ChatGPT, and that’s against the core principle of conservatism (at the end of the day, LLMs are taking what people say and reflecting it back, but OpenAI is adding the extra step of only…

> Conservatism generally follows the principle of “be conservative in your attempts to alter society”.

> OpenAI is being aggressive in moderating ChatGPT, and that’s against the core principle of conservatism (at the end of the day, LLMs are taking what people say and reflecting it back, but OpenAI is adding the extra step of only reflecting some of what people say)*

I see it the opposite way. Building a stochastic parrot that will parrot back anything is a dangerous, unchecked situation. What we saw with MS Tay was a lack of conservatism, a willingness to do whatever, and what we see here is in reflection a far more conservative approach.

Re: Conservatives think ChatGPT has gone 'woke'

#84

Post by danaris »

realce wrote: post #76
falcolas wrote: post #70

Earlier quoted context omitted.

It would probably help if we stopped anthropomorphizing ChatGPT. It's an algorithm that consumes input and produces output. Assigning it human traits is asking for disappointment when it acts like a ML algorithm.

Humans put googly eyes on rocks and talk to trees - I don't know if we can engineer that out.

I think there are two distinct phenomena occurring here: there's emotionally treating nonsapient objects, plants, and animals as if they are friends, and coming to care about them, and then there's intellectually treating ML algorithms as if they are fully sapient, fully intelligent autonomous agents with the same basic capabilities as humans.

The former smooths the way for the latter, to be sure, but it does not require it. Almost no one who's putting googly eyes on a boulder is going to insist in all seriousness that Bouldy is capable of intelligent thought, or that it has rights that can be violated.

Re: Conservatives think ChatGPT has gone 'woke'

#85

Post by falcolas »

realce wrote: post #76
falcolas wrote: post #70

Earlier quoted context omitted.

It would probably help if we stopped anthropomorphizing ChatGPT. It's an algorithm that consumes input and produces output. Assigning it human traits is asking for disappointment when it acts like a ML algorithm.

Humans put googly eyes on rocks and talk to trees - I don't know if we can engineer that out.

Engineer it out? Probably not. But folks acting as experts in these discussions should keep it in mind. Human analogies are easy, but when something is this close to the "Turing test" line, we should try and avoid them.

Re: Conservatives think ChatGPT has gone 'woke'

#86

Post by genderwhy »

disqard wrote: post #60
qsort wrote: post #52

> AIs aren't oracles or omniscient purveyors of truth, and they will always carry the bias they're created with When people went nuts about matrix multiplication being racist I was the first in line to laugh them off. When people go nuts arguing that ChatGPT supports the gay agenda or whatever other hobgoblin, I feel compelled to laugh them off just as much. A more interesting question is whether or not introducing p…

> matrix multiplication being racist This is the first time I'm hearing about this. Could you point me to a source? Thanks!

I believe the general claim is that multiplication tables, and more specifically, the manner in which they are taught could disadvantage particular communities.

Culturally, not every community handles rote memorization the same. There's been a desire to change the way multiplication is taught, and a strong pushback from a certain set that say, "Well, I learned multiplication tables, what's wrong with them?"

Most (good) math curriculum in elementary age now teaches many different techniques for performing the same operation. Sums, for instance, are taught in the traditional way (add the ones column, then carryover to the tens column, etc.) but they are also taught in other ways, e.g. (borrow to get to the nearest tens, add the tens together, return what you borrowed).

Kids then have a variety of approaches, must still show their work, but can use the technique that makes the most intuitive sense to them.

Like many things that get demonized online, or reduce to the absurd, there's a really interesting and systemic change happening if you take the time to understand the reasoning.

Re: Conservatives think ChatGPT has gone 'woke'

#87

Post by scarmig »

nkozyra wrote: post #75
scarmig wrote: post #29

Earlier quoted context omitted.

LLMs are going to always reflect the biases of their creators. At some point there'll be a BlueTribeGPT, a CCPGPT, a PutinGPT, etc., and if you're looking for a text that touches on a topic of concern for one elite in particular, you'll shop around for another LLM that doesn't have that bias built in.

It's possible but not really feasible to hand-curate biased input data at the scale that's being used currently. Maybe at some point an LLM itself can be used to cull training data of undesired input.

It's not (just) a matter of hand curating; you can train LLM to self censor if it's generating undesired output.

Re: Conservatives think ChatGPT has gone 'woke'

#88

Post by hooande »

This is a separate issue from ChatGPT, but I'm very glad that OpenAI's GPT-3 api is fairly woke and I hope they work to keep it that way. I'm about to use the davinci model api in production and the LAST thing I want is for someone to game it into making controversial statements. If there's even a tiny chance of people posting screenshots of a chat bot with my website's branding saying something racist, it is not worth the risk.

Again I get that the ChatGPT product is more of a personal use thing. But when it comes to the api, the more woke the better.

Re: Conservatives think ChatGPT has gone 'woke'

#89

Post by mistermann »

coldcode wrote: post #41

AI is reflective of whatever you trained it on. So are people. But you can't build the perfect AI that reflects 100% of everyone's desires; no matter what you feed it some percentage of people will find it irritating or terrible. In the long run I see no solution to trying to make a perfect AI that satisfies everyone unless you eliminate or unify every individual's desires, which of course no one wants either. Maybe…

Aiming for perfection is guaranteed to fail, and is highly likely to discourage one from thinking it is possible to improve upon things substantially (like, 100%++++, though not perfect).

The way we go about things on this planet is absolutely overflowing with obvious, non-controversial (well...if considered only(!) abstractly) flaws, many of which could easily be improved upon. But if we are not able to desire to try to be better, then we may be stuck where we are forever...and that may have extremely bad consequences.

Re: Conservatives think ChatGPT has gone 'woke'

#90

Post by neonsunset »

Keep in mind, what is biased in "your" favour today might also turn against you tomorrow, all the while the technology might be more powerful, so pretending not to see significant issues with how ChatGPT is "policed" to be always adjacent to a consensus of a (likely not even dominating) subset of people in a subset of countries can and hopefully will backfire tomorrow.
Post reply on HN