Hacker News Reader

Comment by ed312 | Hacker News Reader

I think we're almost at the point where the predictive text bots could all chat with each other in the LinkedIn echo chamber. I look forward to the day where I get a notification from "me" on LinkedIn saying we just had a great interview and start at the new company next Monday.

View on HN

We are absolutely at that point. People are optimizing their content for the algorithm, and using keywords and hashtags so the machine recognizes patterns batter.

Even on a mostly non-algorithmic feed like Reddit, the highest upvoted comments on popular subs always feature an extremely predictable response, like a dad joke, followed by similarly predictable sub-comments written solely to garner upvotes.

> Even on a mostly non-algorithmic feed like Reddit, the highest upvoted comments on popular subs always feature an extremely predictable response, like a dad joke, followed by similarly predictable sub-comments written solely to garner upvotes.

The subreddit simulator is a good proof of this, especially once it was rebased on top of GPT-2. Generally, the closer a subreddit is to the top of the list the more it's content will look like the output of the language model.

Reddit is the very essence of an algorithmic feed. Nunber of upvotes vs downvotes is an algorithm (though a straightforward one), so of course the content will reflect that metric.

Someone needs to make a LinkedIn profile who's content is all GPT-2 generated from other posts and see how many job offers it gets.

Well, a Sea Bream did pretty well.

https://twitter.com/j4n0/status/1125380024733925377

I created a fake almost empty profile to test a previous job's profile scraping and it regularly gets more search hits than my real profile according to the emails.

You can use the built in suggestions in LinkedIn's chat to do exactly this.

Reply on news.ycombinator.com