Very interesting, I was not aware of this shift by Medium. I think platforms like this need to decide even more on a direction, even in the choice of 'who to turn to'. Thank you very much for this issue (P.S. I subscribed!).
The main problem is with the MPP formula itself. By rewarding engagement, you incentivize AI-writing. The previous MPP formula only took into consideration reading time
> they will be bombarded with a flood of low-quality articles
Is that really true? If their recommendation algorithms are any good, users should remain being bombarded with high-quality articles -- AI-written or not, although at this point I would guess the high-quality articles will still be human-written.
If authors flood the system with AI-generated article, in theory Medium should be able to just detect their low quality and they would naturally not be very visible.
The only issue I can imagine to be real is if the flood is so intense Medium's infrastructure would be overwhelmed, but that sounds like a stretch and they could place a limit on the number of articles posted by author by month. A limit of, say, 200 articles/month is still very large for a human author while posing no problem for infrastructure.
Though I don't have knowledge of how Medium's recommendation algorithm works under the hood, I can say this: Every recommendation system has to balance between exploration (recommending new content) and exploitation (recommending popular content). So it will have to at least display new-albeit-poor articles to several users to gather enough signal about them. At scale (i.e., when there is more poorly-written content than high-quality content), this can be enough to dilute the system and degrade the user experience.
On a side note, I don't necessarily think all quality articles will be written by humans. Though I personally prefer to write all my article myself (for now), I think that AI can create good content if humans provide it with the right instructions. The point I'm making is that it should be the readers who decide what is good content (worth paying for) and what is not.
Very interesting, I was not aware of this shift by Medium. I think platforms like this need to decide even more on a direction, even in the choice of 'who to turn to'. Thank you very much for this issue (P.S. I subscribed!).
The main problem is with the MPP formula itself. By rewarding engagement, you incentivize AI-writing. The previous MPP formula only took into consideration reading time
Good summary. This is on top of the struggles Medium has had with the newsletter bloom. They’re saturated an Substack is accelerating for me.
> they will be bombarded with a flood of low-quality articles
Is that really true? If their recommendation algorithms are any good, users should remain being bombarded with high-quality articles -- AI-written or not, although at this point I would guess the high-quality articles will still be human-written.
If authors flood the system with AI-generated article, in theory Medium should be able to just detect their low quality and they would naturally not be very visible.
The only issue I can imagine to be real is if the flood is so intense Medium's infrastructure would be overwhelmed, but that sounds like a stretch and they could place a limit on the number of articles posted by author by month. A limit of, say, 200 articles/month is still very large for a human author while posing no problem for infrastructure.
Though I don't have knowledge of how Medium's recommendation algorithm works under the hood, I can say this: Every recommendation system has to balance between exploration (recommending new content) and exploitation (recommending popular content). So it will have to at least display new-albeit-poor articles to several users to gather enough signal about them. At scale (i.e., when there is more poorly-written content than high-quality content), this can be enough to dilute the system and degrade the user experience.
On a side note, I don't necessarily think all quality articles will be written by humans. Though I personally prefer to write all my article myself (for now), I think that AI can create good content if humans provide it with the right instructions. The point I'm making is that it should be the readers who decide what is good content (worth paying for) and what is not.
Thanks for the update. I’ll leave the partner program and monetize some other way rather than have a platform judge my use of one tool or other.
What’s next: spreadsheets? Word processors? Emails ? The internet? Dictionaries? Pencil and paper?