praash

I believe the present level of datacenter spending will never be justified by actual demand and will crash the global economy.

Qwen 3.8 27b-mlx running locally on Apple M4 48 GB. I only use local LLMs, mostly for getting technical suggestions and code snippet samples. Qwen 3.8 has been the first local model I've tried that has consistently given usable and impressive results.

This also results in a per-string just intonation ;)

Running bash inside zsh continuously destroyed my history for the last 3 years until I figured that HISTFILE was exported and inherited by bash.

My local grocery store goes a bit further, selling € 280-350 laptops with 4 GB RAM and Win11 preinstalled. Absolutely bonkers.

At a glance the repo feels like a meticulous forensic investigation, but I couldn't find any photographs or other information about the alleged drone capture and where all this information is actually sourced from.

I really like your analogy as-is, but LLMs aren't reliable power tools. Asking for external help requires good communication and can be overkill for small tasks.

Hiring a team of subcontractors carries risks unless you babysit the whole project.

Dan, your HN comments are a collection of beautiful essays that is becoming a reference library for healthy discourse. I think it deserves a dedicated website.

> zero-content

Changing the default demo is a reasonable and actionable suggestion.

"Heavy-duty" calls for exaggerated impact and prestige.

"Tony pioneered the famous red-and-green squiggles of Microsoft Word, empowering millions of users with a spell-checking revolution."

The conclusion conflicts oddly with the author's arguments and interests. Attributing LLM usage would actually help the author avoid articles even touched by LLMs, but they indirectly admit being haphazardly dismissive.

Their arguments are mostly addressed by proper, clear attribution. "My sister helped with my homework essay" deserves distrust withour further clarification.

Comparing LLM usage to lying is a fun perspective, but most of the lying happens in attribution. Their moral against lying also seems silly.

I really like this idea of rapidly extensible software, a browser is a nice sandbox for it! Models also seem to be much better at generating programs than "manually" executing a described task.

Lightweight browser extensions generated on demand, now there's a good use case for what they seem to be actually building.

Extending applications without having to launch a full agentic IDE. Macos is already very well equipped with GUI automation tools.

I laughed on the groundbreaking emphasis that you can move and resize the chat window because it's a Mac.

You're up to something, maybe they really have a broken pseudo-window with basic UI interaction hacked on top.