X under Musk’s leadership: Substantial hate and no reduction in inauthentic activity

The paper found that the overall volume of hate speech posted per week increased by 50 per cent after Musk’s takeover compared to the volume before although the volume of total posts had only increased by eight per cent. Similarly, the rate of likes for hate content increased by 70 per cent. This was in contradiction to Musk’s claims about how engagement with hate content had reduced.

Three terms were recorded to have a consistent presence in posts containing hate speech. These were the homophobic “f-slur”, the racist “n-slur” and the transphobic “t-slur”. While other slurs did occur and were recorded in the researchers’ datasets, they didn’t show up enough times to provide a working sample size to gauge their used before and after the purchase of the platform.

Compared to the researchers’ control sample dataset, the hate speech dataset had more original posts (14 per cent compared to 36 per cent), replies (28 per cent compared to 37 per cent) and quotes (three per cent quotes compared to seven per cent quotes). The sample dataset, in contrast, recorded more reposts than the hate speech dataset – 55 per cent to 19 per cent.

Qualitative observation showed that hate posts of a certain category went beyond just insulting people of the targeted identities. For example, both homophobic and transphobic hate posts had a recurring trope claiming that gay and transgender people were mentally ill. However, the transphobic posts exhibited a political angle more often than the other with users stating their support or opposition to politicians based on their stance on transgender rights.

The paper found no considerable change in ‘coordinated account activity’ but did note an increase in the number of bot-like accounts which were active on the site. It also noted an increase in posts about cryptocurrency which are known to contain spam material. The paper adds that large language models have made it easier to produce ‘inauthentic speech’ and thus the researchers have no reason to believe that such posts have reduced after the data collection period.

The paper highlights the need for independent study and evaluation of claims made by social media companies about their platforms, especially since they often lack supporting evidence or data. It also calls for further studies which expand the scope to include other languages and communities on the platform.

The research in the paper only covers ‘overt’ hate speech. Niche internet communities or extremist groups often use ‘covert’ coded language which spreads hate while flying under content moderation systems. The researchers suggest that future work in the area could focus on the changes in such covert hate speech following Musk’s purchase of X and how the trends relate to the prevalence of overt hate speech.

Focus and Factoids by Swadesha Sharma.