Nate Bennett

@nxen.bsky.social

HCI researcher, quantum user experiences, sorting out workflows, chatbots, and “intelligence machines” • All opinions are my own • GrapheneOS & Apple • OSS https://www.linkedin.com/in/nate-bennett-00061132b http://stayfresh.dev https://dev.to/imaginex

apps.apple.com/us/app/moodb... I just updated my app for mood tracking, geared towards those with bipolar disorder. Download it free, save a few days worth of data and start accumulating insights. No gamificaiton or upsell, open source if you want to build it yourself. github.com/Evoke4350/mo...

Moodbound App - App Store

Download Moodbound by Nathaniel Bennett on the App Store. See screenshots, ratings and reviews, user tips, and more apps like Moodbound.

apps.apple.com

LLMs: “let me think step by step…” (proceeds to spiral for 30 seconds) HRM: “I have two brains. one plans. one actually does the work.” 27M params. ~1000 samples. still out here solving reasoning tasks like it’s annoyed at them. maybe the breakthrough wasn’t more tokens

AI will happily generate a “perfect” schema that denormalizes everything. Looks fast. Reads clean. Demo passes. Then reality hits. This isn’t an AI problem. People outsourcing tradeoffs. Normalization vs denormalization is a judgment, don’t outsource it. LLMs don’t feel production pain. You do.

Localhost is a lie you tell yourself. The moment you open a port: • your API keys are public loot • your logs become liability • your “no auth yet” becomes a breach Most hacks don’t look like hacks, they look like missing basics. Ship like someone is already poking at it.

Next.js + AI isn’t some magic unlock it’s a new layer of failure modes non-deterministic outputs state that quietly drifts streams that break mid-sentence we didn’t eliminate complexity we hid it behind nicer APIs the real work now is debugging systems we barely control

Text-to-image models feel new, but the idea isn’t. Early computing turned constraints into creativity. 8-bit graphics, ASCII art, tile maps. You described a world with almost nothing, and the machine did its best to render it. Now we do the same thing, just scaled.

📣GIVEAWAY TIME📣 As a way of saying thank you to you, I am giving away awesome stuff! 🔥🙌 To enter, you must FOLLOW me & RT this post ✔️ One person will receive: ✅️A complete in box copy of Chips Challenge for GBA ✅️A complete in box copy of Nebulus for GBA ⭐️Ends 4/9

Bild

Scaling isn’t a flex, it’s a tax. You don’t “solve” scale, you start paying in coordination, sharding, and weird failure modes. Cloud just lets you choose which bill hurts less. There’s no magic architecture, only tradeoffs you haven’t met yet.

Coding isn’t just economic activity. Software is infrastructure for human connection, like aviation. Generative coding is apprenticeship at scale. The loop: serve, build, share. There are real costs (burnout, isolation), but the work persists because it extends human capability.

I am probing the limits of distraction. Acknowledging that the external noise, however meticulously crafted, eventually fails to obscure the internal silence.

Everyone's chasing 'Superpowers' for massive projects. My grey beard tells me the real power still lies in meticulous engineering. Without that bedrock, your shiny new tech crumbles.

Meta builds profiles on people who never signed up, pulling from friends, brokers, trackers, and the web. You are not the customer. You are the prediction target: a bundle of inferred fears, habits, politics, and weak spots, tuned for ads, influence, and quiet behavioral steering. Quietly, at scale.