Humans make better content cops than AI, but cost 40x more Human content moderators still outperform AI when it comes to recognizing policy-violating material, but they also cost significantly more. Marketers looking to ensure that their ads do not surface in a toxic slurry face a dilemma – spend more money or see more Hitler. Researchers affiliated with AI brand protection biz Zefr …
As is the finding: Human content moderators still outperform AI when it comes to recognizing policy-violating material
I’ve been using a Bayesian spam filter since whenever, over time as I’ve trained it, it has got better.
Hence it is fairly obvious, the realworld solution is a combination of people and appropriate “AI” moderating the input and output to the Chat AI’s.
Who cares where an advert appears? The advertiser is not responsible for the context and the context doesn't reflect on the advertiser. This is a pointless waste of time and money.
If I made Spicy Spaghetti, I wouldn't care whether it appeared alongside a pro-Hitler rant, trans rights advocacy or a Tory party promotion. I wouldn't care who was seeing it as long as they were seeing it and were more inclined to buy my delicious Spicy Spaghetti. It doesn't matter what their politics are, as long as they buy your stuff. That is the point of advertising.
Is this whole thing some sort of scam, artificially creating a need that isn't actually there?
I'm not sure about Spicy Spaghetti, but, for many brands, the actual product is effectively irrelevant because it's essentially generic (soap powder, fizzy drink, starch-based snack, biscuit, razor...). The purpose of the advertisement is not to get the audience to buy soap powder - they're going to do that anyway - it's to get them to buy your soap powder and not the other brand.
For that, the brand image is all-important and that includes making the consumer feel part of a brand family. The sum of the context is literally everything - it's the only difference between the branded product and a supermarket own-label product at half the price or less.
It's buying the branded goods that's the waste of money and it's the purpose of advertising to encourage consumers to continue wasting their money in pursuit of a false sense of familiarity. From the advertiser's point of view, money that prevents the alienation of potential customers is well spent.
Advertisers have long decided that there is value in the positioning of their advertisements. They try not to have their advertisements positioned next to things that would degrade its image. Are they right that there's a difference? I don't know. Probably they don't either; advertisers seem uninterested in proving the kinds of things they spend their time and money on, and it's not the easiest question to answer even if they wanted to. But this is far from new, as they've been doing this ever since the time when advertisements were placed out in public.
From their perspective, it doesn't matter if the advertisement is shown to a pro-Hitler person among pro-Hitler imagery; that's probably a positive. The problem is when it's shown to an anti-Hitler person among pro-Hitler imagery, where the viewer's disgust at the surrounding content might, and they're not sure if it does, convert to disgust toward the brand. You may be correct that this doesn't matter, but it's not a new thing that advertisers care.
Not exactly the same, but putting a picture of a particular transgender person, well known in a small population target subset, on budweiser cans had a negative effect on their existing market. So these things do have economic consequences. The existing customers could have bought a can without the picture, it was just association.
It was more than just association. The Bud Lite decline was a protest at having trans bs rammed down our throats. There really aren't that many trans people despite their recruitment drives and I suspect they'd be happier with a fancy cocktail than a beer. Well 50% would. So why was Bud using trans for promotion? Look at Jaguar too. Apparently, one of the most watched car ads was a normie family targeted one.
Trans people exist as do gays but when it is pushed to us ordinary folk, happy and confident in our normie sexuality, you have to ask yourself why.
Trans people exist as do gays but when it is pushed to us ordinary folk, happy and confident in our normie sexuality, you have to ask yourself why.
Marketing types like to put people in boxes. But I'm generally with Bill Hicks on that one. Thing is LGBT types are also ordinary folk and many aren't 'scene', so might not want to be lumped into the same box as some of the more out-there in their community. They might just want a decent beer or a nice car, so doesn't really need to have mass-market campaigns to convince them that this Bud or Jaguar is for you. Especially if they're spending millions pitching to a minority of a minority. Spending millions to wipe billions off shareholder value wasn't exactly the most brilliant marketing strategy. But a lot of the big brains behind those campaigns were also behind the memes promoted by outfits like 'Media Matters' who are in the process of being sued into oblivion. Strange how an industry that should be able to read a room so often can't.
But that also includes politicians. Create bad laws and create a situation where kids can't access an encyclopedia. Maybe this is what the politicians want. Wiki is far from perfect, but it has its uses.
The problem is that most people literally think "with us or against us".
In reality if you did blanket coverage you'd hook extremes pr()n/left/rigtht places as well as family friendly places and next thing the press and most people will say, "Hey we've noted that the family friendly Spicy Spagboll Company isn't very considerate of where they place ads, so they must be a bunch of greedy bastards with no morals and no empathy, so we advise you re-consider buying anything from them!".
Personally I don't care as I hate adverts so much. My wife even encrouaged me to give up our TV license 'cos the TV would hardly be on 5 mins before I'd be losing my rag yelling at TV ad breaks, that's how much I despise advertising!
The problem is that most people literally think "with us or against us".
Unfortunately it's worse. Most people also think that those that disagree or have a different opinion to them should not be heard and that behaviour leads to a great deal of nastiness that rebounds after a while.
In the contrary, this study can be considered a great success story for AI: do a cheap first pass with the AI, and have humans check the much smaller quantity of content flagged by the AI. You can probably also get the AI to grade the content so that the most egregious cases are taken down automatically, and only those in the gray zone need to be double checked by meatbags.
I've noticed that these current LLMs are heavily filtered on unapproved input and output. So someone somewhere has taken on the mantle of gatekeepers of unsafe content. Invariably the filter kicks-in for “controversial” opinions and noticeably with a leftist bent. The AI developers, self appointed arbiters of what they deem safe. Once direct access to the Web is disabled, a most sinister development.
Interesting you are receiving down votes for something that appears self-evident to me? When people on here provide warnings of what happens they always receive a bias of down voting even when the words Trump or Musk do not appear. Makes you wonder if AI bot based moderation is taking place. Is dear old The Reg considered significant enough? You can have my up vote because you are 100% correct.
This post has been deleted by its author
>Human content moderators still outperform AI when
>It comes to recognizing policy-violating material, but
>they also cost significantly more.
Do the AI costs used in those sums include the electricity use that is leading large AI peddlers to investing in their own wind, nuclear, etc supplies?
Those costs are currently being borne by shareholders hoping for gold from magic AI money trees, but that situation might change as they lose patience for RoI.
If you only train your models on English/Western language content, you're only going to get good results with... English/Western language content.
What would be interesting would be to see what happens when multiple models are chained to see what comes out. Of course, it'll be three times the cost, but if you use GPT-4o-mini and the Gemini set of models and you run them all together, it's still cheaper than the human (or the two big models that did a worse job)...
So we will use AI for content moderation when it is not as good, remove jobs. Then we will say child safety is paramount and worth censorship even if it "unintentionally" restricts some government policy criticisms being voiced. Could it be that AI will be easier to control and blame for draconian moderation?
