How to Touch Grass in the Latent Space

Your feed is convinced we’re either weeks from utopia or hours from extinction. Your brain is not built for daily existential cliffhangers, especially when half of them are engagement bait.

The reveal
"AI doom anxiety" is often a cocktail: real uncertainty, real stakes, and an information environment optimized for emotional arousal. The solution is not "ignore everything." It's building a better pipeline from evidence to action so the discourse can't hijack your attention like a poorly aligned recommender system.

The steel-man (yes, but also no)
Yes: there are legitimate risks worth tracking, and "just relax" is not a policy. No: constant exposure to catastrophic framing is not the same as being informed, and it can degrade judgment—exactly when you need judgment.

Receipts (what we know about doom-scrolling and its correlates)

Look, Satici and colleagues published this "Doom-scrolling Scale" study in 2022 that basically confirms what your therapist already suspected: compulsive bad-news consumption correlates with social media addiction, fear of missing out, and certain personality traits. The key insight? It's measurable behavior, not a moral failing. You can find their work at https://pmc.ncbi.nlm.nih.gov/articles/PMC9580444/ if you want the full methodology.

Meanwhile, NIST dropped their AI Risk Management Framework 1.0 in 2023 (https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf), which is basically a translator for converting "AI risk" into actual organizational controls. Think of it as an antidote to pure panic—boring, systematic, and weirdly reassuring.

And if you want to see what structured capability tracking looks like when it's not designed to make your cortisol spike, check out the UK AI Security Institute's Frontier AI Trends Report (https://aisi.s3.eu-west-2.amazonaws.com/Frontier%2BAI%2BTrends%2BReport%2B-%2BAI%2BSecurity%2BInstitute.pdf). It's an example of how to monitor developments without turning every update into an existential cliffhanger.

AIDOOM's anti-doom-scroll protocol (designed to be boring on purpose)

Here's the thing: you need a system that's basically RLHF for your attention—reward the behaviors you want (useful information) and starve the ones you don't (compulsive refresh).

First, time-box your information intake. Choose a fixed window for "AI news" and stick to it. This prevents the feed from becoming your default mental state, which is what happens when you let the algorithm decide when you're "informed enough."

Second, upgrade your sources. Prefer primary documents and evaluation reports over hot takes. This reduces both variance and emotional manipulation—turns out most scary headlines become a lot less scary when you read the actual paper.

Third, translate everything into mechanisms. When you see a scary claim, rewrite it in terms of prerequisites and bottlenecks. This turns panic into analysis, which is infinitely more useful for actually understanding what's happening.

Fourth, pick one actionable lever. Could be policy, workplace practice, or personal boundaries—doesn't matter as long as it converts helplessness into agency. The goal is to have something concrete you can do with the information.

Finally, exit cleanly. Close the loop with a non-screen activity. This breaks the reinforcement cycle that keeps you coming back for "just one more update."

Disproof conditions (when the anxiety is more justified than the scrolling)

Sometimes you should worry more, not less. The tell is when the evidence pipeline upgrades, not when the vibes intensify.

If we started seeing repeated, independently verified capability jumps in severe-harm domains with weak mitigations, that would justify higher concern because the "risk is abstract" defense would start to shrink. Similarly, if credible incident data showed systemic failures across deployments, the harm would no longer be hypothetical—it would be documented.

The real red flag would be governance breakdowns that prevent slowdowns or halts even after clear tripwire crossings. That's when the incentives-to-catastrophe chain would actually tighten in ways that matter.

Action line
Stay informed at the level of mechanisms and evidence, not dopamine. If your "AI news" intake produces more cortisol than clarity, you're not tracking risk—you're training your own recommender system to keep you scared.

Anyway, back to touching grass—preferably the high-resolution kind.

Related reading

Next up, you might want to check out our doom astrology chart because categorizing claims reduces panic. Or dive into our vibes versus evals piece, since a good evaluation beats a bad feeling every time. And if you're wondering "what now?" after all this, our regulatory capture for dummies explainer might help—because the answer is often policy, not vibes.