Skip to content
Back to posts

How I Use AI to Keep Up With AI (Without Losing My Mind)

Notes by SDG ·

There is simply too much AI news to follow anymore. Between the daily deluge of new models, research papers, product launches, and funding announcements, trying to read everything is a losing game.

Instead of drowning in the noise, I use AI to narrow the field, unpack what actually matters, and preserve what I learn so I can come back to it later. Here is exactly how I run my curation and learning loop.

1. First, I Filter the Noise

To keep my sanity, I built a custom web app called xNoiz. It pulls from the news sites, newsletters, and social feeds I care about and packages them into scheduled digests. I can customize the prompts in the app to pull in exactly (works 90% of the time :)) what I want.

Some days I read it closely; other days I just skim it for a few minutes. The goal isn't to consume every single piece of content—it’s just to find the two or three things actually worth my limited time.

2. Breaking Past the First Layer

When my digest uncovers a complex research paper or a technical announcement, I bring it straight into an LLM. Before I commit to reading a 30-page PDF, I use the model to get my bearings by asking:

  • What is this paper actually claiming?
  • What specific problem is it trying to solve?
  • Why does this matter, and what assumptions is it making?
  • Where could this argument be wrong or weak?
A quick caveat: I never treat the AI's explanation as absolute truth. LLMs are incredibly good at making half-baked ideas sound perfectly logical. I use it strictly to build a basic mental model so I'm not starting from scratch when I dive into the original source.

3. The "Brain Dump" Draft

I don't consider myself a professional writer, so my first drafts are incredibly messy. When I start a blog post, I’m just trying to capture a fleeting idea before it disappears. It’s usually a fragmented, repetitive brain dump of notes, half-formed arguments, and open questions.

Once the raw thoughts are down, I have an LLM clean up the prose. It connects my fragmented ideas, removes the repetition, and translates my chaotic notes into coherent sentences.

This step is mostly for future me. Without it, my notes would only make sense in the exact moment I wrote them. The AI helps preserve my thinking in a format I can actually understand weeks or months down the line. (Of course, I still review the output carefully—smooth writing can easily mask weak thinking).

4. Bringing in a Skeptical Second Model

Once the first model makes my draft look polished, I often hand it over to a different LLM for a cold, unbiased review. I instruct this second model to be critical and look for:

  • Unsupported claims or missing context
  • Weak logic or repetitive points
  • Language that sounds too confident without enough evidence

This "good cop, bad cop" dynamic is crucial. The first model is focused on making the writing sound nice; the second one is there to poke holes in it.

5. Publishing and Simplifying (The Final Test)

Once the piece survives the critique, I publish it on Notes by SDG, a simple webapp that I "vibe coded" to act as my digital archive. Forcing myself to publish is the only way I actually finish my thoughts.

If a story is particularly useful, I'll also share it on LinkedIn. xNoiz has an option to craft custom linkedin posts as starting point, which I then edit.

This final step of summarizing a complex topic into a punchy, short post forces me to strip away the fluff and focus purely on the core message. By the time I share it, I’ve engaged with the idea at least half a dozen times:

[Scan Digest] ➔ [Read Original] ➔ [AI Explanation] ➔ [Brain Dump] ➔ [AI Polish] ➔ [Skeptical Review] ➔ [Simplify for Share]

The Takeaway: It's About Retention, Not Speed

It would be easy to look at this setup and think it's just a system for pumping out content faster. But the real value is much deeper: it helps me move from simply noticing an idea to actually retaining it.

AI handles the busywork of discovery, initial translation, formatting, and proofreading. But the core of the process—deciding what matters, verifying the facts, and applying critical judgment—is still entirely on me. That is how I keep up with AI: not by reading everything, but by building a feedback loop that helps me think.