The For You Page Trains Attention to React. What If It Could Train It to Aim?

The EU accuses TikTok's 'rabbit hole' algorithm of breaching its content rules. That's the headline this week, and it's worth reading in full — regulators say the For You page needs to change how it works in order to protect users' mental health.

You can debate the legal merits for hours. But underneath the regulatory language is a claim most of us already suspect is true: an endlessly optimized feed is very good at capturing attention and very bad at training it. Scrolling teaches your brain to react. It does not teach your brain to aim.

For most people that's an annoyance. For people with amblyopia, it's a missed opportunity on a much bigger scale — because amblyopia isn't a willpower problem. The two eyes send mismatched information to the brain, and so binocular coordination and depth perception never get the practice they need. The result is a lifetime of small hesitations: catching a ball, parking, judging a step, reading a room.

I've been thinking about this less as a policy story and more as an engineering story. Because the interesting question isn't whether feeds should change. It's what an attention-training system would even look like if you built one on purpose — and why the tools in this space have lagged so far behind the attention economy that surrounds them.

The tools have mostly been passive too

Traditional approaches are effective but rarely engaging. An adhesive patch over the stronger eye. Red-blue glasses. Both ask for patience and deliver very little feedback. There's no telemetry, no loop, no signal telling you whether today's session moved anything. If your attention is already being competed for by every app on your phone, "sit still and wait" is a hard sell.

That's the gap we built Amblyotube to fill. Amblyotube, developed by Seven Sports, is recreational and educational software for visual coordination and attention exercises in VR. It runs on Meta Quest, and its core idea is simple: show a different visual experience to each eye while you watch YouTube-style content.

The architectural bet is worth spelling out, because it's the part that took the longest to get right. A normal headset renders one stereo scene to two panels. Amblyotube renders two divergent pipelines — per-eye compositing with independent occlusion, blur, and highlight layers — and keeps them frame-locked so the brain receives a consistent mismatch instead of a flickering one. Any drift between the eyes and the fusion target falls apart.

What actually happens in the headset

Start by choosing your affected eye (left or right) and a severity preset — Mild, Moderate, or Severe — or run the LogMAR vision check. Then pick a video you actually like and launch it in a virtual room, or use Room View AR passthrough so you're still aware of your surroundings.

From there, the app does three things at once:

Dual-eye technology. Your dominant eye gets a gentle, adjustable digital blur via the Dominant Eye Shader. Your lazy eye gets the AI Sharpener (MFBF), which detects human figures in the video and marks them with a bold red outline and a yellow-green highlight, plus a pulsing flicker at a steady rhythm. Optional Wedge Occlusion blacks out part of each eye's view, so each eye sees unique portions of the scene — which cuts down binocular rivalry.

Binocular fusion. A bright magenta focus cue appears in the lazy eye's view, and a subtle neutral grey dot appears in the dominant eye's view. As your tracking improves, the two merge into one cohesive 3D image. That merge is the rep that matters: it's the brain learning to fuse both eyes and process true depth. Think of it as a feedback controller where the error signal is binocular disparity, and the reward is convergence.

Real-world carryover. After a session you take the headset off. No adhesive patches. No red-blue glasses. Just 30-40 minutes a day with content you'd probably have watched anyway.

Why this matters right now

The regulatory fight over the For You page is really a fight about what attention is for. If the answer is "to be harvested," then the incentives will keep pointing that direction. If the answer is "to be trained," you need tools where the user is the one choosing the target.

I keep coming back to a simple design principle: the feed should serve the training loop, not the other way around. That's the difference between an algorithm that optimizes for time-on-app and one that optimizes for a measurable skill. Same delivery surface, opposite objective function.

Amblyotube is one small version of that idea, aimed at a specific audience: people who want to improve their visual coordination and attention and who live with lazy eye or amblyopia.

30-40 minutes recommended daily, maximum 1 hour. Ages 13 and up.

Amblyotube is a supplementary vision training and assistive tool for binocular coordination and depth perception. It is not a diagnosis, treatment, or cure. If you have concerns about your vision, talk to an eye care professional.

Download Amblyotube on the Meta Quest App Store: https://www.meta.com/en-gb/experiences/amblyotube/25906906972338493/