Notes on AI, product design, mental health tech, Rust, and whatever I'm building at the moment.
Filter by tags
26 posts in total
My AirPods Max drained to zero, the reset only blinked amber three times and went dark, and I had already removed them from my Mac and iPad. I spent the day in Bluetooth logs until I found the name of the problem and the most bizarre fix possible.
I opened a clean session as if I were a user and asked how to use Nest. The first command returned command not found. The rest of the day went the same way.
A half-gig file, visually awful, finds the right card as well as the 3.6 GB lossless one. The metric I was using to veto it measured the wrong thing.
Seriously, you used 200 demo cards? I have a client meeting and I still don't have the dataset.
The bytes of a build didn't match an encode done by hand. The tune=still I had been asking for from the start had never been applied, and the manifest said it had.
Lossless means three different things when the source is already JPEG. Separating them changed what I bet on, and the lossless run settled the first bet that same night.
Can you swap 7 GB of Magic card JPEGs for a single file the app opens directly? The first build said yes, and raised two questions I didn't have before.
How to structure schema and JSON-LD to get cited by generative engines like AI Overviews, Perplexity and ChatGPT, using the @graph this blog actually serves as the example.
The reading list I put together on stealing models through APIs, model inversion and gradient leakage, deduplicated and checked, with one line on what each source does.
In February I proposed sorting images by similarity before encoding the dataset in AV1 and estimated 50 to 80 times compression. In September I tested it on Magic cards, and the idea only survived where there were near-duplicate images.