Nick Tiedemann

Karpathy just said the people who don't use LLMs are already losing. he spent 4 minutes explaining why smart people are still going to fall behind. Not only the people who refuse AI, but also those who think signing up for Claude counts as using it. here's what it looks like for most people right now: > ask Claude to rewrite an email > ask Claude to summarize something > close the tab that's not wrangling Claude. that's paying $20/month for spell check. Karpathy's point isn't that LLMs are powerful. everyone knows that. his point is that knowing how to use them is the actual skill gap and most people are nowhere near closing it. Claude can be your research analyst, your writing editor, your salary negotiation coach, your financial reviewer, your 30-day curriculum builder. all of that is in your $20/month subscription. right now. today. most people will see this tweet, agree, and go back to asking Claude to fix a sentence. the article below covers 20 prompts across every area of your life. not productivity hacks. actual use cases that change how you work and how you think. the model is not the bottleneck. knowing what to ask is.

The only Chinese model out right now that holds a candle to GPT-5.5 or Opus 4.7 is Qwen 3.6 Max/Pro. Change my mind.

A few months ago I built a rather pricey dual-GPU computer for local AI inferencing. For months, whenever I’d actually run inference on it, the system would hard lock. This is a known issue on certain combos of kernel, driver, and hardware, so I tried many different combinations with zero luck. It happened again yesterday, and I stepped away in absolute frustration. In a park near my house, I realized something: it might not be software. When I got back, I ran inference task on my GPUs and did a memory error check. Bingo. One of the GPUs (the non-display one) had over a million ECC memory errors, meaning at least one of that card’s physical memory chips was bad. Not all GPUs have ECC memory, so I was fortunate to be able to check for this. It is a good reminder that software solutions can’t truly fix hardware problems.​​​​​​​​​​​​​​​​ Also go touch some grass because it’s good for you.

@willeastcott @playcanvas The capability to do this has been around for a long time, but the real estate consumer wasn’t ready for it. Interesting sociological shifts here.

Videogame technology is disrupting the real estate sector 🏡 📷 Scan a house 🪄 Train a 3D Gaussian splat 🌐 Publish to the web with @playcanvas Buyers can speedrun through property listings! 🏃 Try it for yourself on SuperSplat 🔗👇

@icreatelife cruft

Prove you work with AI with just one phrase

Your Claude can be 10x smarter. For free. This skill uses subagents ("monkeys") to attack problems from novel angles, judge the results, and preserve the input-output chain. It builds a knowledge base of how good results are produced, so it can do better in the future. github.com/nstied/monkeys…

A key product discipline.

I recently flipped Codex to fast mode and left it there. Burns quota quick but worth it for me. I've got like 10+ half-done projects, I forget what is even happening. Fast mode burns my tokens, then I touch grass, no sitting and waiting like a dumb rock. Even parallel subagents, Ralph loops, etc. have their limitations. In terms of project calendar time, this actually works out for me. It lets me spend my time in physically healthier ways. AI is still slow and checking work takes too long. We're nowhere near filling the hunger for smarter, faster AI.

AI is not creative enough. "Brainstorming" skills are always one-offs. So I made a skill that fixes that. Monkeys with Finger Paints generates crazy ideas, judges them, and preserves chain-of-custody forever. It gets smarter. It leans good taste over time. Github link below