Cutting AI noise: 13 high-signal sources and how to consume them

Executive overview

Most AI content is noise. The challenge isn't finding more sources — it's knowing which 20% to keep.

The key filter is purpose: know whether you're consuming for entertainment, education, or utility. This video optimises for utility — practical AI you can apply today or in the near future.

Two consumption modes (audio and reading) map to specific sources. Tactics like playback speed, sampling with AI, and a read-later tool reduce time spent without sacrificing signal.

Cut 80% of your sources by filtering ruthlessly for signal-to-noise ratio relative to your specific goal.

Audio sources

• Latent Space (podcast) — focuses on applied AI research and building; hosted by Swix, who has unusually broad visibility into the Silicon Valley AI ecosystem
• No Priors — covers startups and new AI products; a compact way to track what's shipping
• Dwarkesh Patel — not AI-only, but extracts answers from Anthropic, OpenAI, and Google guests that don't surface elsewhere; prioritise his deep-dive AI interviews
• AI Explained (YouTube) — covers new research and models with genuine analysis, not just blog post recaps
• AI Jason (YouTube) — focused on vibe coding; good source for net-new techniques in that space
• AI Engineer (YouTube) — summit talks only, not regular content; production-grade use cases from research labs and top AI startups
• Sequoia AI conference — one recent release rated unusually high signal-to-noise; not worth following for general content

Reading sources

• AI newsletter (aggregator) — daily; read only the exec summary at the top, which takes ~30 seconds; aggregates signal from Twitter, Discord, and Reddit
• Simon Willison's blog — read selectively (~15% of posts); most valuable when a new model drops — he runs consistent practical tests rather than benchmarks
• Eugene Yan's newsletter — deep dives on specific topics (RAG, evals, prompting); published every 2–3 weeks; read ~2 in 10 issues

Community and social

• Latent Space Discord, #ai-general-news-and-chat — aggregates the best AI chatter from X without requiring you to be on X; skim every 2–3 days, ~20 minutes catch-up
• Two X accounts worth following: one focused on vibe coding tips, one on AI news with genuine editorial perspective (neither named explicitly in the video)

Consumption tactics

• Audio over reading — consume podcasts during walks, workouts, cooking, cleaning, driving; treat dead time as learning time
• Playback speed — human speech is low information density; start at 1.25x and work up; ~1.75x is the practical ceiling before comprehension drops
• Sampling before committing — for long podcasts (2–3 hrs), paste the transcript into o3 or Gemini 2.5 Pro and ask for the most compelling discussion points and why they're novel; listen only to those that pass the filter
• Read-later tool (Pocket) — save links when you encounter them; read during dedicated reading time (before sleep, while stretching), not in the moment
• Skim, don't read — for newsletters, read the exec summary or intro only; skip the rest unless a specific piece is directly relevant