"LinkedIn isn't banning all posts generated by artificial intelligence. Some, she concedes, actually have some value. Others, though? Those need to go. They won't be vanishing anytime soon, however. As the company refines the tools that will hunt out the offending posts, it will be rolling things out slowly-and it could be several months before all users see less slop in their feeds."
"The new systems will target three types of AI content: generic AI-generated posts and comments, attention-bait videos, and automation tools that create AI content. Hunting the robots LinkedIn isn't offering a lot of in-depth details about how it plans to scrub this content, but Lorenzetti says the company is using an "AI solving AI" approach."
"Newly built systems will parse posts and determine which of those offer original thinking and which lack substance. The systems are designed to learn over time, using the engagement patterns of users and identifying language that adds perspective versus simply regurgitating existing ideas. Human editors will be involved as well, labeling thousands of posts as original or generic to help teach the AI which posts to flag and which to leave alone."
"Similarly, the company is putting together a list of markers that are common among low-quality, AI-composed comments to purge those from the system in the future. Identifiers such as word patterns and the volume of comments are key to that hunt. (An AI tool, for instance, can compose and post s"
LinkedIn plans to cut back on low-quality AI posts that distract users from valuable professional content. The company will not ban all AI-generated posts, since some can provide value, but generic and low-substance content will be targeted for removal. The rollout will be gradual, with improvements taking several months to reach all users. New systems will focus on three categories: generic AI-generated posts and comments, attention-bait videos, and automation tools that generate AI content. The approach uses “AI solving AI,” where systems evaluate originality and substance, learn from user engagement patterns, and distinguish perspective from regurgitated ideas. Human editors will label thousands of posts to train the systems, and marker lists will help detect low-quality AI comments using patterns and comment volume.
Read at Fast Company
Unable to calculate read time
Collection
[
|
...
]