Of course someone that sells cloud computing would publish this drivel. My local qwen3.8 27b system doing valuable work night after night disagrees.

Yup agreed, I've ran 1 billion tokens through with >3000 steps and multi-hour autonomous output with nothing more than a 5090, Deepseek Harness, and Qwen 3.8 27b Quasar NVFP4. You get 260k context, ~150 token/s continuous performance, and full agentic capabilities.

I wouldn't call $0.2/$1.2 "ridiculously cheap"

There is no "small team" pushing 100PRs/day, agents or not.

From my limited tests, Qwen has better reasoning which is a bummer because Muse Glimmer is literally twice as fast.

I really don’t understand how people can even consider LLMs to be anything more than an extremely sophisticated auto-complete. Logic thinking itself emerges from completing patterns. Nothing more to it than that.

I don't understand how people can claim to know that, given we don't know what consciousness is

As a software developer, I absolutely despise loud keyboard sounds that I didn’t produce myself. In an office setting, fast typists with loud mechanical keyboards are the absolute worst (thank God for AirPods)!

The fast typists who don't know how to use a mechanical keyboard, are the worst, that's true.

I was running Immich for a while until the iPhone client app on my wife’s phone completely stopped syncing photos. So I ended up vibe coding my own photo management software in .NET using PostgreSql/Redis and React front end (PWA). Has face recognition too (used the same models Immich uses from huggingface). Works perfectly, photo library scanning/face recognition/thumbnail creation/etc performance is WAY better than Immich (uses .NET background jobs and lots of parallelism and hardware acceleration on my Mac mini server). Turns out if you only care for the thing to work on your own gear you can optimize the code for it quite a bit. It took 2 weekends and Claude code. And with tailscale, it’s hosted on my Mac mini at home and accessible from anywhere through https. I have around 40k+ photos+ phone videos, and the server is a base Mac mini previous gen (8GB ram). Oh, and forgot to add, the app supports downloading/moving photos from iCloud through the undocumented CloudKit APIs behind the iCloud.com web app, complete with 2fa.