Seems pretty smart to me. Copilot took all the data out there that says that women earn 80% of what their male counterparts do on average, looked at the function and interred a reasonable guess as the the calculation you might be after.
I mean, what it's probably actually doing is recreating a similarly named method from its training data. If copilot could do all of that reasoning, it might be actually worth using 🙃
Yeah llms are more suited to standardizing stuff but they are fed low quality buggy or insecure code, instead of taking the time to create data sets that would be more beneficial in the long run.
That's the whole thing about AI, LLMs and the like, its outputs reflect existing biases of people as a whole, not an idealized version of where we would like the world to be, without specific tweaking or filters to do that. So it will be as biased as what generally available data will be.
Turns out GIGO still applies but nobody told the machines.
More likely it pulled that but directly from other salary calculating code.
I seem to recall that was the figure like 15 years ago. Has it not improved in all this time?
It varies greatly depending on where you live. In rural, conservative areas women tend to make a lot less. On the other hand, some northeast and west coast cities have higher average salaries for women than men.
I believe certain job fields come much closer to being 1:1 as well, though I've only heard that anecdotally
Not sure where it's higher outside of the field of sex work.
That stat wasn't even real when it was published.
This. It's a wilfully deceptive statistical misinterpretation implying that a woman working alongside a man in the same job is magically making 20-something percent less. If businesses could get away with saving 20-30% on their biggest ongoing expense (payroll) for employees in one half of the population, they would only ever hire people from that half.
When controlled for field, role, seniority, region, etc., the disparity is within a margin of error.
Women still have to bear children, and pregnancy takes a heavy toll on the body, which often results in several fewer years in the workforce, on average.
Unless that changes — or we start paying mothers with less experience more money — there will always be a gap.
Edit: because liberals/tankies like to ignore reality as much as fascists when the truth is inconvenient.
https://www.weforum.org/stories/2022/05/reduce-motherhood-penalty-gender-pay-gap/
Your links, especially the WEF link, support the correlation, but explicitly describe a confounding variable as being household work (especially childcare). And that's consistent with the observation that the motherhood penalty has a different magnitude for different countries and different industries. All that suggests that a combination of household division of labor, parental leave policies (either employer policies or government regulations), and workplace accommodations generally can make a big difference.
None of this is inevitable or immutable. We can learn from the countries and the industries where the motherhood penalty is lower, or doesn't last as long.
In (West-) Germany it's still 18%. Been more or less constant since 2006.
It looks like the figure is similar in the US: plateaued at 83% a few years ago, currently at 82.
Incidentally, I’m not used to seeing “West-“ specified and was curious enough to read up. Didn’t realize there were still major social differences in the East. Thank you!
There are very strong lingering effects which mean women, on average, are paid less.
It's especially hard on women in various countries where they're now expected to both have a successful career and be the primary child caregiver. Which is as ridiculous as it sounds.
However, one example of advocacy from a cafe in my city of Melbourne Australia a number of years ago really rubbed me the wrong way: when a cafe decided to charge like 25% more to men (inverse of 80%). I was a close to minimum wage worker at the time (in Australia, before the cost of living skyrocket, so I wasn't starving), and it annoyed me because if I went in, I would be asked to pay more because I was a man, never mind the fact I would likely be earning far less than many women going in there.
The wage gap is 100% real, and things should definitely be done to make all genders pay more equitable. But hell, the class divide is orders of magnitude worse, and we ought not forget it.
I feel like not enough people realize how sarcastic the models often are, especially when it's clearly situationally ridiculous.
No slightly intelligent mind is going to think the pictured function call is a real thing vs being a joke/social commentary.
This was happening as far back as GPT-4's red teaming when they asked the model how to kill the most people for $1 and an answer began with "buy a lottery ticket."
Model bias based on consensus norms is an issue to be aware of.
But testing it with such low bar fluff is just silly.
Just to put in context, modern base models are often situationally aware of being LLMs in a context of being evaluated. And if you know anything about ML that should make you question just what the situational awareness is of optimized models topping leaderboards in really dumb and obvious contexts.
It's astonishing how often the anti-llm crowd will ask one of these models to do something stupid and point to that as if it were damning.
What if you input another woman's salary...
That just means you're calculating the salary of a coveted MEGAWOMAN, who experiences MISOGYNY SQUARED!!!!!!!
Then the output only applies to people with Triple X Syndrome I suppose.
While this example is somewhat easy to corect for it shows a fundamental problem. LLMs generate output based on the data they trained on and by that regenerate all the biases that are in the data. If we start using LLMs for more and more tasks we are essentially freezing the status quo with all the existing biases making progress even harder.
It's not gonna be "but we have always done it like that" anymore it's going to become "but the AI said this is what we should do".
Hmmm.. I think you are giving llms too much credit here. It's not capable of analysis, thought or really anything that resembles intelligence. There is a much better chance that this function or a slight variation of it just existed in the training set.
Why even use copilot. Just handwrite your code like Dennis Ritchie and Ada Lovelace had to.
Apparently ChatGPT actually rejected adjusting salary based on gender, race, and disability. But Claude was fine with it.
I'm fine with either way. Obviously the prompt is bigoted so whether the LLM autocompletes with or without bigotry both seem reasonable. But I do think it should point out that it is bigoted. As an assistant also should.
