The Loathing Is Real. So Is the Usage.

The Adoption Curve That Resentment Cannot Reverse

The social media comparison has become a fixture of AI criticism, but it is usually deployed as a warning. What the Bluesky commenter captured in a single post is that it functions better as a structural prediction ^[17]: if the arc matches, then public loathing is not a counterweight to adoption — it is a feature of how adoption works at scale. The people most harmed by a technology are often the people most embedded in it, because opting out carries costs that resentment does not offset. This was true of Facebook and Instagram. It is already true of AI tools in hiring, academic, and professional contexts.

When the Biggest Critics Are the Heaviest Users

The generational split that has emerged around AI is not a split between users and non-users — it is a split inside the user population. Young people whose AI use is most intensive are simultaneously the cohort registering the sharpest opposition to it. The mechanism is not hard to locate: being told that AI will destroy the job market and also being required to demonstrate AI proficiency to enter it produces a specific kind of trapped resentment. The same double bind shaped social media adoption among millennials — platforms that corroded their social experience while becoming the mandatory venue for professional visibility. The pattern is not a paradox; it is a labor dynamic.

Sam Altman's conference admission that AI is not very popular in the U.S. right now reads differently against this backdrop. It is not the statement of a CEO worried about declining usage — it is the statement of a CEO who has separated popularity from adoption and found the separation reassuring.

The Regulatory Analogy and Its Limits

Bernie Sanders's question — whether anyone believes the past two decades of social media were good for children's mental health ^[18] — lands with rhetorical force and historical irony. The same question, in more or less the same form, was asked about social media in 2011, 2014, 2017, and every subsequent year the evidence accumulated. The New Republic's argument that AI could take social media's harms and turbocharge them ^[19] is analytically sound, but it is also the argument that was made about algorithmic recommendation engines in 2016. Regulatory guardrails arrived late and narrow. The platforms did not wait.

The framing that treats this as primarily a moderation or safety problem misidentifies the structural failure of the social media era. The problem was not that regulators failed to stop bad actors — it was that no accountable public alternative emerged to compete with platforms whose incentives were constitutionally misaligned with user welfare. The same architectural choice is being made again, at speed.

The Argument That Social Media Critics Never Won

The most consequential line in Anil Dash's widely circulated Bluesky post ^[17] is the one that got least attention: the argument for "accountable, public alternatives, not just regulating the bad actors." This is the argument the social media reform movement never successfully operationalized. It came too late, attracted too little capital relative to the incumbents, and found no political coalition willing to fund it before the platforms had become too large to displace. The AI version of that argument is being made now, in public, by people who watched the social media version fail. Whether it arrives at a different conclusion depends not on whether it is correct — it is — but on whether it is being made at a point in the cycle where it can still change infrastructure rather than just critique it. The social media arc suggests it is not, and the people making the argument know it.