Yes please! I still have an iPhone 12 Mini, had to go to the Apple store to change the battery. Of course, they broke the screen while replacing it, so what was supposed to take an hour took the entire day instead.

If they made it easier to replace them we could just replace them ourselves

Agree in general but I do regret buying a cordless vacuum cleaner, next one will have a cable for sure. Only ~3 years after buying it, the battery only lasts for 30 minutes and require 1+ hour to charge, takes entire day just to vacuum the house now, including the forced breaks.

Happy to answer questions if any arise :)

I'm currently compiling together a webpage that will display the exact sessions I had, on all platforms and with all the details, hopefully will add some more insights into the workflow too.

I grabbed all the .zip files as soon as I saw they were available, thinking something like that might happen. Should I re-upload them somewhere, would that be helpful?

I was just looking for something like that yesterday! Fun timing :)

Really liked the video, especially after the 1 hour mark :) Love the honest tone, and explanation of your background!

One thing I felt wasn't mentioned, when talked about everything electrified in the home, was about water and sewage, which one typically forgets about 😅

Great as always regardless!

Agreed, "one skilled human + one agent" builds better software than "one human + hundreds of agents", at least today.

I made that case using more words here: emsh.cat/one-human-on...

Knowing how stupid these people are, she probably sees it as a compliment.

s/almost away/almost always

Even without on-device AI, smartphone notifications almost away go through remote services, often unencrypted. Best to not have notifications enabled at all if you're really paranoid.

Browsers should have build-in markdown parser! I'd love to not have to ship a Markdown renderer in every frontend, and be able to do "window.mdToHTML" or whatever instead!

Nah, 96GB of VRAM gives me enough to run models that are good enough even for professional work today, and doesn't have horrible prefill performance. More important the harness at this point, most "downloable" weights today are more similar in performance than any outstanders, besides GPT-OSS-120b

I'm like 74% sure the future of programming with LLMs is mostly with local LLMs running on our hardware

Cheers, probably less useful, I run LLMs on my GPU in my desktop, and probably wanna avoid slow Macs for that :) This is mainly about being able to build and run E2E tests on actual macOS hardware basically.

Yeah, only for CI, and I already have Windows and Linux setup :)

Or, hear me out, fun leaks from governments around the world, as git repositories, published as blobs on atproto :)

I dunno, I kinda feel like if the model produced it, you already lost. Better way is to figure out the way so they don't write bad code in the first place. They get it wrong? Probably your prompt wasn't good enough, iterate on it and restart from beginning

If no human is reading the code, no human is writing the code, why even have linting rules? They were created for humans to have an easier time reading, writing and reviewing code, so we don't have to nitpick about these things and argue about it

But if the machines does it all, why do we have it?