Twitter's Open-Sourced Algorithm Is "Dishonest," Expert Says

Twitter recently open-sourced much of its recommendation algorithm, as Elon Musk had promised, but critics immediately dubbed the release as “incomplete,” while a former Twitter exec told Kylie Robison of Data Sheet that the release was “dishonest.”

Photo Insert: a former Twitter engineer alleged that the company published only around a fifth of the algorithm.

This is so, Robison added, because Twitter recommends tweets based on a user’s data as well as the code it has now made public, so “in order to open source the algorithm you need to open source the training set, which is impossible for Twitter to do.”

Affirm product manager Aakash Gupta, who authored a viral thread analyzing the code, wrote that a former Twitter engineer told him the company published only around a fifth of the algorithm.

However, as made clear by Gupta and others, we can divine some genuinely interesting information from the release (apart from the fact that Twitter collects metrics segmenting users into the categories of Democrat, Republican, “power user,” and “Elon,” which explains why the egotistical maniac could use his inhouse and “outhouse” stans to dethrone Barack Obama as the most popular Twitter account holder).

Likes provide a bigger boost than retweets do and way more than replies. Adding links actually hinders the quest for a boost, unless the user gets a lot of engagement. Mutes and unfollows really hurt a user’s engagement score. Posts get heavily penalized for misspellings or made-up words.

WEEKLY FEATURE : MVP Group Keeps Lights On During Pandemic