OpenAI used Kenyan workers on less than $2 per hour to make ChatGPT less toxic

To put it in perspective, the average income in Kenya is around $1.25 an hour and tbh (as a Kenyan) I really can't see it as a bad thing. A lot of people here live in abject poverty, they live in situations you can't really begin to imagine. So any sort of help coming our way is good.

&quot;The contracts stated that OpenAI would pay an hourly rate of $12.50 to Sama for the work, which was between six and nine times the amount Sama employees on the project were taking home per hour.&quot;<p>OpenAI paid $12 an hour, it&#x27;s Sama that paid them $2 an hour. People should be upset at Sama, not OpenAI. I&#x27;m sure you could find Americans willing to label data for $12 an hour.<p>From OpenAI&#x27;s perspective, they likely thought they were doing good by going with Sama, since Sama&#x27;s mission is to provide work to people in Africa. From their website: &quot;Since our founding, we’ve pioneered new technologies and business methods that harness the power of markets for social good, meaningfully improving employment and income outcomes for those with the greatest barriers to formal work.&quot;<p>This is a story about Sama exploiting cheap labor in Africa, if that&#x27;s what you want to call it, rather than OpenAI specifically aiming to do that.

Almost more interesting in that article I found that Sama, the Kenyan company, was also asked to collect CP and brutal&#x2F;gore images for OpenAI. After delivering 1400 images, Sama cancelled the contract as this was even too much for them. OpenAI then talked about &quot;miscommunication&quot; and they actually did not really want those CP images. Well, they made its own category for it and asked Sama to collect images of several other categories, but somehow it was lost in communication that for one of theses categories, no images should be collected. Because they are illegal. OpenAI swears they never opened the images they received (and paid for).

Am Kenyan in Tech&#x2F;science. For the nature and the volume of work, the wages sighted here are grossly low. In addition, it appears that Sama was paid in US dollars and negotiated a great contract with OpenAI only to then exploit Kenyans and legally hind under the umbrella of low wages. Shame shame shame. This is so distressing &quot; ... All of the four employees interviewed by TIME described being mentally scarred by the work. Although they were entitled to attend sessions with “wellness” counselors, all four said these sessions were unhelpful and rare due to high demands to be more productive at work....&quot;

2 USD is currently ~248 KES, which is actually a reasonable wage for unskilled labor in Kenya. If you can do this work for eight hours a day, for roughly twenty days a month, then you&#x27;re making ~40,000 KES per month. That&#x27;s a very good salary for a huge portion of the population in Kenya.<p>Source: living and working in Kenya since 2007.

Headline implies there’s something negative here, but I’m not seeing it. Was OpenAI supposed to hire high-skill workers in developed countries to do this work?

I suspect this may not have gone down the way the article is reporting.<p>OpenAi paid Sama $12&#x2F;hr for the work and they in turn paid their employees $2&#x2F;hr.<p>Yet because workers were reportedly affected by the content Sama cancelled the contract?<p>I think we all know they would just replace the $2&#x2F;hr workers.<p>It doesn’t check out

I grew up in Kenya. Low level work pays about 10,000 Ksh for cleaners, construction&#x2F;warehouse workers e.t.c<p>That&#x27;s ~$80 USD&#x2F;month. At 160hrs&#x2F;month, that is $0.5&#x2F;hr.<p>Not ideal, but $2&#x2F;hr is an upgrade for many.

The terrible situation in Kenya (and all other poor countries) is created by the predatory loans provided by IMF (and the countries that copy those types of loans).<p>It&#x27;s unproductive to talk about the hundreds of billions of dollars of slavery work that companies do until we have an answer for the trillions of dollars of toxic debt created by these huge organizations.

I’m surprised no one here is challenging the assumption that ChatGPT needs to be “less toxic”.

When are we going to learn?<p>Uber did Greyball and nothing happened. Its corporate charter should&#x27;ve been revoked and the executives jailed. A very clear message has been sent to the tech companies by this: do whatever you want, the law doesn&#x27;t apply to you.<p>OpenAI did this and nothing happened.<p>At the next pandemic, some sort of this generative AI will be used to write the anti scientific articles and woe upon to you then. This time it was bad but that will be a thousand times worse.<p>Can&#x27;t find who said OpenAI basically released a zero day on human cognition without a patch. Outlaw it already until we find one. This is no joke here.

I feel like comments here are too focused on the $2 per hour aspect and not enough on the emotional abuse that this labour entails. Sure, maybe it&#x27;s an &quot;average wage&quot;, but the people performing this work are suffering mental trauma because for hours upon hours they are subjected to reading stories and viewing images of sexual abuse, assault, gore. The people performing this sort of extremely difficult emotional labour should be taken care of. It is disgusting for OpenAI to pay so little and at the same time have such little respect for humanity and decency.

So? This is good salary locally I believe.

Sad to see the viewpoints of so many people framing this as &quot;helping Kenya&quot; or &quot;They took the job, so it&#x27;s worth it to them&quot;. I can assure you OpenAI did not have a charity meeting and say, &quot;what poor, impoverished country can we lift up today?&quot; It was &quot;What country has limited labor laws, no unions, no red tape, and the cheapest labor we can find?&quot; 99.99% of corporations put profit above people. That&#x27;s why we still &quot;benefit&quot; from child labor and sweatshops, because we prefer cheap crap and money over the betterment of humanity. The same arguments given here were also given by coal miners in 1840&#x27;s Europe[1],[2] when employing women and children, along the lines of &quot;oh the children love it, they even get to see horses!&quot; (and indeed, the children did show up to work every day, so why make a fuss?)<p>[1]<a href="https:&#x2F;&#x2F;www.calderdale.gov.uk&#x2F;wtw&#x2F;search&#x2F;controlservlet?PageId=Detail&amp;DocId=100908&amp;PageNo=2" rel="nofollow">https:&#x2F;&#x2F;www.calderdale.gov.uk&#x2F;wtw&#x2F;search&#x2F;controlservlet?Page...</a><p>[2]<a href="https:&#x2F;&#x2F;api.parliament.uk&#x2F;historic-hansard&#x2F;commons&#x2F;1842&#x2F;jun&#x2F;07&#x2F;employment-of-women-and-children-in" rel="nofollow">https:&#x2F;&#x2F;api.parliament.uk&#x2F;historic-hansard&#x2F;commons&#x2F;1842&#x2F;jun&#x2F;...</a>

Everyone is focusing on the salary, but have you read the article?<p>It&#x27;s legitimately problematic that there is a dark, vile essence of this being abstracted away -- the shiny, attractive, safe-for-advertisers front, vs the violent and disgusting &quot;back end&quot; with manual labourers sifting through the muck.<p>It feels like it&#x27;s a consequence of indiscriminate ingestion of datasets. Like, did OpenAI really have to point their data hoovers at 4chan?

Selfish bleeding hearts are such an annoyance. They never give you good jobs. They always take away the jobs you can get.<p>This makes sense, though. If you work for cheap for someone who is willing to pay you, both of you are happy. The person who is unhappy is a third party who you are harming by preventing him from charging more.<p>So that guy will invent justifications: no job should pay this little, they don&#x27;t know better, etc etc<p>But from the fact that these people never get you a better job and always take away a good job, you can tell what they want: to stop you from outcompeting them.<p>Place it in obvious terms like that and it&#x27;s clear who is trying to harm you and who is a feasible partner.

I grew up in the era of peak-Adbusters, and ate up the narrative on exploitative sweatshops where people in 3rd world countries are forced to make my shoes on near slavery wages.<p>Fast forward a decade, and my now mother-in-law described how she worked in one of these shops, literally making Nikes, and it was an amazing opportunity for her at the time, and helped get her out of poverty. This caused an instant perspective shift.<p>Thats not to say that we shouldn&#x27;t be on the lookout for exploitation, and pushing for better working conditions in general, but the truth (as always) is more nuanced than bad guys and good guys.

&gt; [...] OpenAI’s outsourcing partner in Kenya was Sama, a San Francisco-based firm that employs workers in Kenya [...]<p>Does this qualify as nominative determinism, given OpenAI founder Sam Altman&#x27;s username is... sama?

People supporting this have a similar mentality to Arabian Gulf countries that treat laborers poorly: “we pay you so that makes us generous, shut up”. Of course, this behavior is unacceptable and needs to be condemned. Imo Using people’s conditions in poverty as a way to severely exploit them is like pimping out a starving child and wanting credit for sharing some of the earnings. There is nobody to protect the child and he’s desperate, but that doesn’t make the pimp correct.

AI has always prompted ethical discussions but now that big ones are publicly available we society needs to speed run actually implementing major new ethical decisions.<p>Yes ChatGPT needs to be “not toxic”.<p>The company leadership believes this and is investing heavily researching and designing and implementing as many safe guards as possible.<p>They know if they don’t do a good job here it will be a PR and support disaster and ultimate effect the bottom line.<p>No serious customers will buy any AI product if there’s a chance it spews sex, hate speech or violence, etc. to their end users.<p>At least for now, some humans somewhere in the chain will have to review toxic content to help train.<p>It sounds particularly depressing to me that a $50B company can outsource this to Kenya for $200k. But I don’t know who else or how much money would make it better.<p>Totally hypothetical but I bet some of the handsomely paid US engineers have a ton of policies in place to eliminate exposure to toxic content. But also when they do see some they have a much better foundation in life for it not triggering major mental health problems.<p>What we really need to know is what OpenAIs whole portfolio of approaches, budgets, staff and outsourcing looks like.<p>In the meantime kudos to Time for exposing one piece for public discourse.

I don&#x27;t have a strong opinion on this myself, but I am curious how the general sentiment of &quot;that&#x27;s a good wage for Kenya and this is fine&quot; might change if they were hiring developers there instead.

I&#x27;ve been living in Kenya for the last 7 years. ChatGPT is exploiting workers. Yes many live in poverty, especially in Nairobi, but one &#x27;good&#x27; meal on the side of the road is about $1. To put it in better perspective 1 dozen eggs cost $1.80. Even in slums there is a cost for living in something barely more than a shanty house, without running water and&#x2F;or proper toilets.<p>My vote is ChatGPT is being exploitative, what they are paying should be an hourly rate. They are taking time from people with education and who speak at least two languages. Just because you can pay people less, does not mean you should.

people who are complaining about the wage being too low relative to western wages are<p>1. thinking in words instead of things<p>2. not looking at the alternative which is would kenyan&#x27;s themselves be rather paid $2 or get nothing at all<p>i&#x27;m african not kenyan and i&#x27;m glad these job opportunities exist.<p>the headline is fake outrage from sjw westerners drinking lattes that cost 5 quid &#x2F; 7 bucks

Maybe I need more coffee, but wouldn&#x27;t it be easier to not add text from certain parts of the internet, or make scrapping bots with controls to avoid those areas? I understand it would still grab some unwanted things, but it seems it would be much easier to deal with. Unless of course, this is basically the entirety of the internet and I just refuse to see it as such.<p>There is a section that mentioned &quot;An online search for the text reveals that it originated from an online erotica site&quot;. If you dont want sexually explicit text, then maybe dont include text from literary erotica sites in the first place? Or is it like advertising where companies keep passing around the same chunk of data and continue adding to it? Still, I would think making or paying for better(smarter) scrapping bots would be cheaper.

Please note the cognitive dissonance if you believe both that this is wrong and that remote workers&#x27; pay should be adjusted based on their location.

Forget pay, I&#x27;m more interested in the quality of the training data this produces. If you have Kenyans rating what constitutes hate speech or obscenity, you are going to end up with an AI that reflects Kenyan mores. If you pay peanuts, you&#x27;re going to get low-effort evaluations that miss subtly. Even with a handbook and instructions, this is unavoidable to a certain degree.

I wonder how much this would have costed if they used mechanical Turk instead.

well, if local workers are ready to work for just terms and it&#x27;s legal...

Talk about a story with a political agenda.<p>Just seems like a plain old political outrage clickbait article with the words ChatGPT bolted in to make it topical. The kind which I&#x27;ve apparently developed a new dismissive reflex towards.

Looks like the median salary in Kenya is about $3.50 an hour ($7k&#x2F;year)

A Kenyan working 40 hours a week for two dollars an hour is making twice the GDP per capita. A US citizen working 40 hours a week for US minimum wage is making 1&#x2F;5 of the GDP per capita

Wages in San Francisco are way higher than Alabama, too.

Toxic article from Time about ChatGPT using Kenyan Workers to make ChatGPT less toxic. We went full circle here.

The company OpenAI used to manage workers in Kenya, Sama, has come up before<p><a href="https:&#x2F;&#x2F;time.com&#x2F;6147458&#x2F;facebook-africa-content-moderation-employee-treatment&#x2F;" rel="nofollow">https:&#x2F;&#x2F;time.com&#x2F;6147458&#x2F;facebook-africa-content-moderation-...</a>

As a devil&#x27;s advocate, would all people against $2 salary be also support a call on maximum salary? E.g. A software engineer with 2 years experience cannot get more than $25&#x2F;hr because that is the just global average decided?

Why didn&#x27;t they ask the &quot;AI&quot; itself to filter the results? Isn&#x27;t AI going to offer medical and legal advice soon? How will it do such things when it can&#x27;t tell what is child abuse? Rhetorical question, at least for me.

I find this kind of hard to believe. Having Africans remove &quot;toxic&quot; content is absurd, because Africans tend to be incredibly misogynistic and homophobic with an incredible amount of hatred and vitriol towards &quot;the west&quot; (and now Ukraine).<p>Right now, go onto Facebook, search for Ukraine and open the first 10 posts. You will see that at least 70% of all comments are racist and from Africans.<p>Now search for &quot;gay murder&quot; and you will find that 90% of the comments are from Africans agreeing with the murder and blaming the gay victim.<p>Now do the same search for &quot;priest beheaded&quot; .. then &quot;priest burned&quot;.<p>Having Kenyans train ChatGPT surely means they brought their religious and homophobic biases. This dataset must be garbage.

Bear in mind Time magazine is presumably not the greatest fan of large language models, since they&#x27;ll be churning out Time magazine articles by the 100s, given another year of development.

Do projects like this still exist? The overhead to develop an AI that doesn&#x27;t actually understand anything must be quite high at this point.

It’s funny that @sama’s company is working with a company named Sama. It also confused me when reading some of the comments at first.

Alternative headline: OpenAI hired workers and paid them money to help with RLHF training for their GPT-3.5 model.

It can both be true that the wage is pretty decent for Kenya, and also that OpenAI should have paid more.

Have to wonder what&#x27;s the definition of toxic.<p>Anything that&#x27;s not sufficiently politically correct?

So whatever harm the Kenyans have saved us from, doesn’t that cancel out by exposing them to it?

This has got to be the most dumbass news article related to chatgpt I have come across

To all the likely incredibly highly paid people here (by world standards) commenting on the poor Kenyan wages.<p>Rather than having vacuous, facile discussions on social media that achieve absolutely nothing other than signal how virtuous you are, why dont you donate a significant chunk of your wages to an African charity?

~US$2 is the minimum wage in Brazil, Turkey, Russia, Chile, and a few others.

Just curious does anyone found ChatGPT contain Kenyan ethics bias

Don&#x27;t worry about the $2 because later on they used everyone else for free to see if we can trick ChatGPT into sudo mode, describing how to produce methamphetamine, or producing increasingly racist jokes.

it seems either way kenya will stay poor?<p>i think the solution is to either not colonize or not make such a mess.<p>We should ask the Brits what their idea was.

As a result, they traumatised a whole bunch of people (and their progeny) who will continue to live the trauma much longer than OpenAI will exist.

This is good fo Kenya, Thank you guys.

At least Kenyan workers get $2 an hour ... I haven&#x27;t seen a single penny for the hours of my life spent on recaptchas.

Two dollar an hour censors. I miss the days when the pope would personally review scientists&#x27; correspondence.

Tech journos having to scrape the bottom of the barrel for the Silicon Valley outrage du jour.

Awful journalism

I see so many comments from self-anointed do-gooders that I just have to explain something from my lived experience.<p>I&#x27;m Nigerian. I graduated high school at 15 and because my family didn&#x27;t have the funds for uni, I got a job in what was essentially a sweatshop factory with extra steps. I got paid 12k Naira ($16) per month. After all, I had was a secondary school education and I couldn&#x27;t bargain for better.<p>Roughly one year later, I got a job writing for a content mill where I was paid $1 for 100 words. I can&#x27;t describe my happiness when I got my first $50 for 5,000 words. All earned while sitting at a desk, typing away on a broken laptop, using unreliable internet.<p>To a social justice warrior in SF or NY, it might have been exploitation, but to me an African teenager with no relevant experience or higher education, it was the first step I needed to pull myself up by my bootstraps.<p>I levered up and today, I make the same or a bit more than quite a few freelance writers in the West.<p>Point is, it&#x27;s okay to feel bad for these Africans; but if you really do, you will support free trade - that&#x27;s the most successful instrument for lifting humans out of poverty.<p>Without this &quot;starvation-level&quot; gig like some of you may call it, these workers would settle for something less.<p>Today, Nigeria&#x27;s minimum wage is around 30k Naira, yet many Nigerians don&#x27;t even make that much. On the other hand, a worker making $2 per hour <i>8 hours </i> 5 days will net $80 per week or 59k Naira at the current market rates.<p>If you complain someone is exploiting me, please give me a better job - or let me take my least worst option.

Meanwhile, your shirt and pants were made by Indian children working for less than $2 per hour.

Can’t we rename OpenAI to ClosedAI?

To be fair, I find the censorship makes it more toxic.

So much comments saying it&#x27;s good for kenyans while most of us would not review gore and toxic content all day long for our own local minimum wage. This is clear abuse of cheap labor, and also a good way to not provide psychological support that would be required in our own country.

Could they afford to pay US minimum wage to workers, and they’re merely pocketing the difference?<p>Or are they wholly unable to afford US minimum wages and are papering over that inability through Kenyan labor?

I find the following paragraph to be quite funny.<p>&quot;To get those labels, OpenAI sent tens of thousands of snippets of text to an outsourcing firm in Kenya, beginning in November 2021. Much of that text appeared to have been pulled from the darkest recesses of the internet. Some of it described situations in graphic detail like child sexual abuse, bestiality, murder, suicide, torture, self harm, and incest.&quot;

Just going to say it....These companies are run by criminals....all of them are always connected to crime. it&#x27;s a fact. not an opinion. OpenAI has blood money. It&#x27;s blood diamonds but information.....same thing....just modernized and remote....