Filter for posts: Musk wants to publish Twitter algorithm


Elon Musk plans to release Twitter's algorithm that filters the content users see in the “For You” tab. The code is scheduled to be made available as open source on March 31st. As is so often the case, Musk barely left Twitter's existing infrastructure a good hair when making the announcement.

In view of the code quality, the transparency in this case was initially “incredibly embarrassing”, he announced on Twitter. In his view, the algorithm is overly complex and would not be fully understood even internally. Users would therefore discover many “stupid things”. Patches should come as soon as problems have been identified. He expects faster improvements from being open.

Recommended external content Twitter

At this point there is external content from Twitter that complements the article and is recommended by the editors. It can be loaded and hidden again with one click.

Load Twitter embeds I consent to Twitter embeds being loaded. Personal data can be transmitted to Twitter. More on this in the data protection declaration.

Load Twitter Embeds Privacy Policy

In addition to improving the existing algorithm, Musk says Twitter is currently working on a simplified approach to recommending suitable tweets to users. This code is also to be made available as open source, but development is still ongoing.

Recommended posts as construction site

In any case, according to Musk, Twitter wants trust win. A necessary step, because since the takeover, the recommendation algorithm has been considered a construction site. Complaints have been accumulating about odd posts appearing in users' feeds.

Musk has long described the recommendations as a nuisance. According to media reports, he felt disadvantaged internally, so in February he asked for a code adjustment. The result: His posts flooded the “For you” tab of the users regardless of their relevance.

In addition to the “For you” tab, Twitter offers “Following” nor the classic view, which sorts Twitter posts chronologically.