In a nutshell, the ANES data shows:
📉 Social media use is shrinking; engagement collapsing
💥 Twitter/X posting has moved ~70 POINTS to the right
🧩 Platforms are splintering
🔊 Fewer people are talking — but those still talking are more politically extreme
A side-by-side infographic titled "The Great Information Shift: Pyramid vs. Network," with a dark blue banner header and subtitle about the move from 20th-century top-down control to 21st-century decentralised information flow. Left half — "20th Century: Hierarchical Pyramid": A pyramid made of blue blocks, narrowing to a single block at the top crowned with a microphone and a crown icon, labelled "Top-Down Authority." The middle layer is labelled "Institutional Gatekeepers (Centralized Verification)." At the base, a crowd of grey silhouetted figures represents the "Passive Public (One-Way Flow)." Blue arrows point only downward from the top through the gatekeepers to the public, emphasising the unidirectional nature of information. Right half — "21st Century: Flat Network": A web of interconnected orange nodes and lines forming a dense mesh. The nodes are all roughly equal in size and represent different actors — icons depict news outlets, social media likes, megaphones, cameras, screens, and people — labelled collectively as "Equal Nodes: Institutions, Influencers, Citizens." At the centre sits a gear-like icon labelled "Algorithmic Amplification." Orange arrows radiate outward in all directions between nodes. The bottom label reads "Multidirectional Exchange (Real-time Feedback)," and a side label identifies participants as "Producer-Consumers." The overall argument is that information flow has shifted from a hierarchical, one-way broadcast model to a flat, many-to-many networked model where everyone both produces and consumes content, mediated by algorithmic amplification rather than institutional gatekeeping.
That architecture has now been replaced. Social media platforms curate what most people see. Their logic isn't verification, it's engagement. Whatever holds attention gets amplified, regardless of truth or consequence.
"are you mogging me right now?"
The NSF 2027 budget has noted that they will close out the Social, Behavioral, and Economic Science Program (SBE). This is not a good thing. nsf-gov-resources.nsf.gov/files/FY-202...
Interesting.. I wonder how much of that is online norms leaking offline. Much of the volume looks at hate communities specifically but the rhetoric has spread far beyond niche platforms. Seeing it IRL is awful and suggests a breakdown of face-to-face norms that were once thought to be pretty durable
If you're interested, an edited volume came out last year that focuses on, in part, the entertainment value of hate. it's available for free here www.taylorfrancis.com/books/oa-edi... (full disclosure, I co-wrote one of the chapters)
*ARIZONA CHARGES KALSHI WITH OPERATING ILLEGAL GAMBLING BUSINESS
Yah, there are productive and counterproductive ways to use this tech, which sometimes be difficult to sort out. But either way it's not a great idea to become dependent on something that may not exist in the future or will undergo major enshittification via ads or $$$
more broadly the risk to universities of institutionalizing this tech is that tools change and often get more extractive through advertising and/or monetization. students who 'learn to use AI' may find themselves dependent on tools that become either expensive or nonexistent in the form they learned
See also: social media APIs for academic research
haven't you heard? the world is ruled by.. a rounding error
once overheard at a conference "if Jews and Hindus combine forces we'll be 1.2 billion strong"
And it was doing a lot of this even before Musk www.pnas.org/doi/10.1073/...
Article: The political effects of X’s feed algorithm Abstract: Feed algorithms are widely suspected to influence political attitudes. However, previous evidence from switching off the algorithm on Meta platforms found no political effects1. Here we present results from a 2023 field experiment on Elon Musk’s platform X shedding light on this puzzle. We assigned active US-based users randomly to either an algorithmic or a chronological feed for 7 weeks, measuring political attitudes and online behaviour. Switching from a chronological to an algorithmic feed increased engagement and shifted political opinion towards more conservative positions, particularly regarding policy priorities, perceptions of criminal investigations into Donald Trump and views on the war in Ukraine. In contrast, switching from the algorithmic to the chronological feed had no comparable effects. Neither switching the algorithm on nor switching it off significantly affected affective polarization or self-reported partisanship. To investigate the mechanism, we analysed users’ feed content and behaviour. We found that the algorithm promotes conservative content and demotes posts by traditional media. Exposure to algorithmic content leads users to follow conservative political activist accounts, which they continue to follow even after switching off the algorithm, helping explain the asymmetry in effects. These results suggest that initial exposure to X’s algorithm has persistent effects on users’ current political attitudes and account-following behaviour, even in the absence of a detectable effect on partisanship.
Figure 2. ITT estimates of feed-setting changes on engagement and political attitudes. ITT effect estimates of switching the algorithm on and off (in s.d.). Left, effect of moving from the chronological to the algorithmic feed for users initially on the chronological feed. Right, effect of moving in the opposite direction for users initially on the algorithmic feed. For each outcome, the results of two specifications are reported. Blue, unconditional estimates with robust s.e., controlling only for the initial feed setting and, where applicable, pre-treatment outcome levels. Orange: conditional estimates, controlling for pre-treatment covariates using GRFs; 90% and 95% CIs are reported. Numerical effect sizes and P values correspond to the conditional estimates (all tests are two-sided). The unit of observation is respondent. From top to bottom, sample sizes are n = 4,965, n = 3,337, n = 4,965, n = 4,965, n = 4,596, n = 4,596 and n = 4,850. Tests are described in Methods. Supplementary Information Table 2.16 reports the exact numerical point estimates, s.e., CIs and sample sizes for every specification. All outcomes are standardized. Additional results are presented in Supplementary Information section 2. PCA, first principal component from principal component analysis.
X's algorithm is in fact doing what you think it's doing. www.nature.com/articles/s41...
interesting to me that they did not break down the 'moderate' category further, given the large sample size. A 0 to 12.5 hour/week range captures a pretty broad set of behaviors.
Colleges are in trouble for many reasons but a big one is that they started believing that students are customers and not products.
thanks! on it..
I’m teaching a class on digital traces in the spring and would love to check this out if you’re willing to share!
This nicely reinforces findings from our recent @pnas.org piece in which we argue “that industry players like Meta make significant investments into long-term research streams … to absolve their platforms of responsibility for adverse effects on society or individuals.”
www.pnas.org/doi/10.1073/...
One, possibly small, factor that has us in this situation is that there is no norm of using private messaging apps among the (non-immigrant) American population. I suspect that this will have to change in the coming years if anything like a coordinated resistance movement is to emerge out of sight..
The transformer was invented in Google. RLHF was not invented in industry labs, but came to prominence in OpenAI and DeepMind. I took 5 of the most influential papers (black dots) and visualized their references. Blue dots are papers that acknowledge federal funding (DARPA, NSF).
@theloftcinema.bsky.social
bsky.app/profile/bost...
Something I teach my students is that reading is a rhythm. They need to build a space-time where the rhythm of longer form reading is available to them. Generally this begins by having them inventory the various kinds of noise (cellphones, screens, work/care/social commitments) in their lives.
🔥 "You’d ask a bot for a summary and forget what it told you, then proceed with your day, unchanged by words you did not read and ideas you did not consider."
That little cc moment is satisfying, though
