Show HN: Derail – Fight back against algorithmic addiction with LLMs

The internet has become adversarial. Youtube contains a treasure trove of knowledge and information but is also filled with time sucking shorts and slop. Knowledge is locked within these trillion dollar platforms managed by PM's who unlock million dollar stock grants for every minute of increased engagement. The web is now incentivized to keep you dissociated and scrolling.

The effects on your brain are worse than you think. Researchers found that endless feed and infinite scrolling disrupts memory encoding. The more you use social media, the more everyday memory slips. Feed prevents construction of self. After a while I came up with a counterintuitive realization: the only way to regain control is to surrender control. Human willpower is finite, and it should not be spent on fighting distractions. LLMs, however, are perfectly suited for this task.

So I built a chrome extension called Derail. You tell Derail what your goal is, and it uses open source models to evaluate every page's content and determines if it should be blocked or allowed through. Distractions are curtailed at the earliest stages possible before you get sucked into the mindless doom scrolling loop.

Both the goal prompt and blocking prompt are customizable, so each person can fine tune the blocking behavior to their level of strictness. Are you conducting economic research and want access to newspaper and financial videos only on Youtube? Done. Are you learning math and only want access to math lectures on Youtube? That's also achievable.

Letting models act as gatekeeper to the internet sounds extreme. But against trillion-dollar companies built to devour your attention, it's barely sufficient.

It's an early prototype right now, and classification isn't perfect, but it has made a positive impact on my life. I'm releasing this tool and paying for inference out of pocket, because this is an issue I truly care about.

If this resonates with you too, I would love for you to try it out and give me any and all feedback about how it can be improved.