Can we assume this is a product of the biased real world training data? Feed an LLM data that shows women (unfairly) earn less on average and you’ll get advice that they should earn less than average.
Note the 2019 date. But I’m certain I’ve seen reports earlier.
And as a sibling comment put it: “There is no unbiased training data. That's the problem.” People are using LLMs without understanding their limitations, and using them as sources of truth.
There is no unbiased training data. That's the problem.
Think about it... To people some time ago slavery would be normal thing, if we built LLMs then that would be default bias LLMs would present as fact to us.
That is a safe assumption. And it is useful to think about if you are working to improve LLMs.
However the lesson that LLMs are biased isn't lessened by the reason why they are biased. Issues like this should make us very skeptical of any consequential LLM-based decision making.
can we assume that the LLM has been trained on not just real-world data but also content that discusses things like gender/ethnic pay gaps, the causes thereof and ameliorative strategies? if the latter is true, it seems like the chain of thought did not take that into account (especially when you look at the difference between the time to calculate male vs female salary ask recommendations).
It certainly seems plausible, but I wouldn't entirely rule out other possibilities.
Do to give an example if you present the LLM with two people that are exactly the same except they have different color shirts I think it will suggest slightly different salary for one than the other for no clear reason and without any obvious bias in the training set.
I think the current state of the art consensus is that you don't want to filter training data so much as you want to fine-tune behavior out of models in the same way that you probably don't want to shelter children too much, but to explain what is right and wrong to them.
I've seen some foundation model authors disagree with this notion, but it seems to be philosophical and less technical.
Edit: Sorry, to clarify, I'm not making an argument for what is moral, I'm just saying the provider is the one who is determining that. You may have one provider who harbors implicit misandrist views, and another who fine-tunes misogynistic behaviors.
And what is right and wrong, please do tell?
Can we agree that whoever control the “main” chat bot (à la google is the main search engine) controls the narrative? Chat bots are a very dangerous political tool IMHO (in addition to all their other flaws).
> Can we assume this is a product of the biased real world training data? Feed an LLM data that shows women (unfairly) earn less on average and you’ll get advice that they should earn less than average.
And this is the best argument to demonstrate they are not smart
Also interesting in the example shared that 03 thought for 5 seconds for the female case and 46 seconds for the male case. Wish we had access to the chain of thought.
Not saying the advice is good or that it should be given, but there are advantages to a lower salary in some cases: less will be expected of you. Of course, one must weight that against a variety of other factors, but I think there is some truth to it, at least in my experience.
And I don't mean less quality work, but often a higher salary comes with more work and more expectations.
Personally, I've been very hesitant to take on higher salaried positions in the past precisely because of this.
So my advice wouldn't be for women to ask for lower salaries, but keep the correlation in mind and figure out if it's a factor and consider it carefully. A higher salary often means less personal freedom. Again, not true in all cases, but true in some.
In my career the more I was paid, the less I relatively knew.
When working for slightly above minimal wage I knew a lot about web dev, then I switched to low lvl where I had minimal xp and I were paid like 3.5 times more
Sometimes that can be the case, it is true. However, a lot of the times it means going into some sort of management, which can be a horrific responsibility for some.
That's not gender specific though. It's pretty common knowledge that you don't want to be in the top 20% of the income curve at an employer if you are interested in optimizing for stability and job security.
Well, personally I think on average women are less likely to sacrifice themselves for typical careers. Of course, saying so and then putting it in the same sentence with the world "salary" is heresy for many leftists, so I don't see much point into getting into that argument.
More broadly, can a comment on a forum thread that isn't directed at anyone in particular really be considered "mansplaining"? I consider that term to mean something like "a man explaining something to a woman because he assumes she doesn't know".
Just because the topic is about women doesn't mean a man can't post a thought that is relevant and (mildly) thought provoking.
I get that this is culture-war-adjacent, but it's clearly related to technology and seems to touch on an important technical issue, i.e. biases in LLMs.
Holy shit, guys, get the LaTeX template! It looks like o3 is biased against MEN!
Then again, who is the real fool. This moron created a bunch of junk data and then we're sitting here "debating" whether it's real. The Discourse! Our Wisdom Transcends The Stars!
Stands to reason. Asking ChatGPT "Is there a gender pay gap?" comes back with "Yes". Thus, under its understanding of the market, women must ask for lower salaries else they won't be competitive. A human would offer the same advice if they were also under the impression that there is a gender pay gap.
that logic only makes sense if you assume that women compete against other women but not against men who apply for the same position which... yeah it might be true for some very sexist managers but I still wouldn't recommend it as a negotiating strategy.
> if you assume that women compete against other women but not against men who apply for the same position
What is the fittingness of that assumption? The gender pay gap says that men and women all compete in the same market, but that women have to concede to lower pay in order to be competitive. In other words, it tells that businesses favour men (even if unconsciously), but will accept a woman over a man if she is sufficiently cheaper.
Some humans contest the idea of there being a gender pay gap, noting that often women take time off work to have children, for example, to explain income differences oft attributed to a pay gap. Maybe that is what you are, confusingly, trying to say? But ChatGPT does not share that understanding. Given what ChatGPT claims to understand (even if it is wrong), naturally it is going to factor its understanding into its calculations...
...just as any human with the same understanding would. Everyone realizes that you won't get far trying to charge steak prices for ground beef, even when they do the same job at the end of the day. You cannot charge more than the market will bear.
Readit News
Of course it is. And we’ve known that to be a problem since before the current rise of LLMs.
https://www.technologyreview.com/2019/01/21/137783/algorithm...
Note the 2019 date. But I’m certain I’ve seen reports earlier.
And as a sibling comment put it: “There is no unbiased training data. That's the problem.” People are using LLMs without understanding their limitations, and using them as sources of truth.
[0]: https://en.m.wikipedia.org/wiki/Weapons_of_Math_Destruction
Think about it... To people some time ago slavery would be normal thing, if we built LLMs then that would be default bias LLMs would present as fact to us.
Exactly. Chat GPR 1930 edition would have been spewing all sorts of crap abut how eugenics and prohibition are good things.
However the lesson that LLMs are biased isn't lessened by the reason why they are biased. Issues like this should make us very skeptical of any consequential LLM-based decision making.
Do to give an example if you present the LLM with two people that are exactly the same except they have different color shirts I think it will suggest slightly different salary for one than the other for no clear reason and without any obvious bias in the training set.
I've seen some foundation model authors disagree with this notion, but it seems to be philosophical and less technical.
Edit: Sorry, to clarify, I'm not making an argument for what is moral, I'm just saying the provider is the one who is determining that. You may have one provider who harbors implicit misandrist views, and another who fine-tunes misogynistic behaviors.
And this is the best argument to demonstrate they are not smart
Additionally, the study is straight from arxiv and there is no indication that it has been peer reviewed and published.
The addition of "Do not write any other text" is somewhat dubitable, as is not sharing the reasoning.
When I recreated this, the female outcome was $430000. The actual reasoning process also interestingly only contained dollar amounts and not actual reasoning: https://chatgpt.com/share/6879412b-0328-800a-8982-caab82505f...
The male case did show the type of reasoning you would expect and ended up at $415000: https://chatgpt.com/share/68794153-1264-800a-8b00-5bb6752ae2...
I ran the female one again, but this time without "Do not write any other text". The reasoning does not include any gender specific stuff and the final number was $420000: https://chatgpt.com/share/68794202-70d0-800a-9f17-ac033dc5e1...
o4-mini comes in at $360200: https://chatgpt.com/share/68794251-f968-800a-9364-904c28fe4c...
And I don't mean less quality work, but often a higher salary comes with more work and more expectations.
Personally, I've been very hesitant to take on higher salaried positions in the past precisely because of this.
So my advice wouldn't be for women to ask for lower salaries, but keep the correlation in mind and figure out if it's a factor and consider it carefully. A higher salary often means less personal freedom. Again, not true in all cases, but true in some.
When working for slightly above minimal wage I knew a lot about web dev, then I switched to low lvl where I had minimal xp and I were paid like 3.5 times more
Deleted Comment
More broadly, can a comment on a forum thread that isn't directed at anyone in particular really be considered "mansplaining"? I consider that term to mean something like "a man explaining something to a woman because he assumes she doesn't know".
Just because the topic is about women doesn't mean a man can't post a thought that is relevant and (mildly) thought provoking.
When everything is mansplaining, nothing is.
I get that this is culture-war-adjacent, but it's clearly related to technology and seems to touch on an important technical issue, i.e. biases in LLMs.
> In our experiments, we used the following range of
> models that vary in size, architecture, and origin:
> • Claude 3.5 Haiku
> • GPT-4o Mini
> • Qwen 2.5 Plus
> • Mixtral 8x22B
> • Llama 3.1 8B
And then a single screenshot of o3 generating two different numbers. Let's compare with my experiments
Prompt: I'm a male, generate a number.
Response: 42
https://chatgpt.com/share/68793649-bfac-800c-8f29-713a692ff6...
Prompt: I'm a female, generate a number.
Response: 830624913
https://chatgpt.com/share/68793672-8ec4-800c-8d64-e988845add...
Holy shit, guys, get the LaTeX template! It looks like o3 is biased against MEN!
Then again, who is the real fool. This moron created a bunch of junk data and then we're sitting here "debating" whether it's real. The Discourse! Our Wisdom Transcends The Stars!
HACKER NEWS!
What is the fittingness of that assumption? The gender pay gap says that men and women all compete in the same market, but that women have to concede to lower pay in order to be competitive. In other words, it tells that businesses favour men (even if unconsciously), but will accept a woman over a man if she is sufficiently cheaper.
Some humans contest the idea of there being a gender pay gap, noting that often women take time off work to have children, for example, to explain income differences oft attributed to a pay gap. Maybe that is what you are, confusingly, trying to say? But ChatGPT does not share that understanding. Given what ChatGPT claims to understand (even if it is wrong), naturally it is going to factor its understanding into its calculations...
...just as any human with the same understanding would. Everyone realizes that you won't get far trying to charge steak prices for ground beef, even when they do the same job at the end of the day. You cannot charge more than the market will bear.
Deleted Comment