141–150 of 155 posts
Earlier quoted context omitted.
Social stuff on GitHub is weird. I have a couple of old colleagues that fill up most of my feed daily with stars to repos.
I’ve liked how they recently roll these up, like “sixstringtheory starred such-and-such and 5 other repos.” When they switch from stars to hearts, we’ll know the end is nigh.
Earlier quoted context omitted.
> Now sure, you can complain that the algo isn’t showing you what you personally want - its trained on the entire viewer population, not on your personal tastes. So most algos let you personalize the feed. But without an algo…there’s just no hope. I think that the common complain is not "it isn't showing what I personally want" VS "it isn't showing what people generally want" but rather that the algorithm is often no…
Most of the comments here are incredibly dystopian - automation is bad, rehire the editor you've replaced at 75% of the salary, stop pushing engagement, what brings the company most money, megacorporation pushing slushy mass.... Jesus!! Man, didn't know recsys invokes such vitriol. There are dozens of news apps on smart TVs these days. Under the covers, it's pretty much the same deal - in go the raw feeds. Editors al…
yes, corporations do whatever makes the most money. pointing that out has nothing to do with emotions. welcome to the real world.
I look forward to the quasi-spam slurry of infiniscroll feed crap once growth hackers start trying to game the algorithm to get more traction for their project. I don’t think you can introduce a recommendation algorithm without it having a negative effect on the content it’s supposed to aid discovery of. Pre-recsys, the content is made for human consumption, but once you add the recsys, the AI itself becomes part of…
> I don’t think you can introduce a recommendation algorithm without it having a negative effect on the content it’s supposed to aid discovery of. This feels like it ought to be a corollary to Goodhart's Law ("When a measure becomes a target, it ceases to be a good measure"), or perhaps a specialized application thereof. https://en.wikipedia.org/wiki/Goodhart%27s_law
So whether you have a recommendation system or a chronological feed or whatever other way you want to display info—there is always that implicit Goodhartish: measure → target → behavior change.
You can't escape it.
The only question is how you want to harness it—how you can bring out the best in people (with their consent ideally!) and mitigate the negative impacts of Goodharting.
(Shameless plug, I'm working on this, e.g. https://techpolicy.press/can-algorithmic-recommendation-syst... )
Earlier quoted context omitted.
> I'm not writing and debugging 10,000 lines of C just to farm Github stars. And neither would anybody else. That's the problem. (i.e. they'll do something else besides write and debug code to farm stars) Let's say I determine (or at least believe ) that the GitHub algorithm prefers projects with a README with lots of images and emoji, MIT-licensed, and lots of forks. Obviously, there are already some really great pr…
Create 100 different npm projects that should be one bit of functionality but ship every single method as a different github repo and update all of them continuously with new content in their README.md and such in order to suggest to the AI that you're super busy in maintenance. Then abandon all of them once you land the FAANG job you're looking for.
I look forward to the quasi-spam slurry of infiniscroll feed crap once growth hackers start trying to game the algorithm to get more traction for their project. I don’t think you can introduce a recommendation algorithm without it having a negative effect on the content it’s supposed to aid discovery of. Pre-recsys, the content is made for human consumption, but once you add the recsys, the AI itself becomes part of…
> I don’t think you can introduce a recommendation algorithm without it having a negative effect on the content it’s supposed to aid discovery of. This feels like it ought to be a corollary to Goodhart's Law ("When a measure becomes a target, it ceases to be a good measure"), or perhaps a specialized application thereof. https://en.wikipedia.org/wiki/Goodhart%27s_law
The size of transistors was a good target for a very long time, like 60 years.
The height of basketball players.
Muscle mass, profile, or volume.
The profit margins of a company.
Scores on a standardized math test.