nanobuilds

The data comes straight from Google's Ad Center (myadcenter.google.com). Google shows you the interest categories and brands they've assigned to your profile. I automated scraping that page daily for each account during the experiment.

MirrorMask actually does exactly what you're describing. It scrapes your Ad Center profile before and after each session and shows you the diff. You can watch interests appear and disappear over time. The dashboard tracks your profile changes across sessions.

The work machine differences are interesting. Google uses OS, browser language, and the language of content you consume as profiling signals, not just searches. If your work browsing is mostly English-language sites, Google may be slotting those accounts into a different market entirely. Same IP, but different OS telemetry, different default browser settings, different behavioral fingerprint.

Would be interesting to check myadcenter.google.com on each machine to see how different the profiles actually are.

Three fresh Google accounts on different US residential proxies. Two browsed with specific personas (fisherman, fitness). The third did nothing for five days.

Google was more aggressive than I expected. 17 new ad interests from a single session. By Day 1 it was already removing interests and replacing them. Not adding to your profile. Rewriting it. The control didn't move once in five days.

I built a Mac app (MirrorMask) that does this against your real profiles. Happy to answer questions about the experiment or methodology.

Three fresh Google accounts on different US residential proxies. Two browsed with specific personas (fisherman, fitness). The third did nothing for five days.

Google was more aggressive than I expected. 17 new ad interests from a single session. By Day 1 it was already removing interests and replacing them. Not adding to your profile. Rewriting it. The control didn't move once in five days.

I built a Mac app (MirrorMask) that does this against your real profiles. Happy to answer questions about the methodology.

It's interesting how terminal apps are increasing in popularity after decades of desktop and web apps. I wonder if it's the talk to the chat AI that's making people more used to asking a prompt screen or if it's the simplicity and lack of bloat.

Was expecting a higher jump for plumbers. Cashiers were already suffering with the self-checkouts for years.

AI tools are great but ultimately it's about taste and intent (and fun for games) that will hopefully lead to quality floating to the top.

The experience of the internet would be so much more interesting if the search engines unearthed rare blogs or writing from small creators and bloggers that thought things through or shared original ideas.

It did seem we had that for a while and now everything funnels back to a handful of big platforms.

Maybe as AI swallows the data of the entire web, it would start to look for these small sites, small creators, and rare personal content to keep itself interesting and we'll see more of them?

Your human context also needs compacting at some point. After hours of working with an LLM, your prompts tend to become less detailed, you tend to trust the LLM more, and it's easier to go down a solution that is not necessarily the best one. It becomes more of a brute forcing LLM assisted "solve this issue flow". What's funny is that it sometimes feels that the LLM itself is exhausted as well as the human and then the context compacting makes it even worse.

It's like with regular non-llm assisted coding. Sometimes you gotta sleep on it and make a new /plan with a fresh direction.

One can hope that vibecoded apps will eventually be vibe-maintained with agents trained specifically for the kind of novel and weird bugs ai-coding tends to bring up. These tools will hopefully also get better at identifying security risks created but previous generations of ai models. 6 months is a long time in the life of a vibe coded app.

The mobile is much than TV because of the enforcement loop and algorithm turning that attention into scams, ai generated slop, and harmful content that is more and more difficult to identify. With tv I suppose you can change the channel, with reels, the next 20 reels might continue to show you similar things or follow folks on other platforms as well.

If you're upset about this bill:

- Call your MP (find yours at ourcommons.ca). - Back organisations that fight back (OpenMedia and CCLA have killed surveillance bills in the past - Submit written opposition.

The Cannabis Act angle is interesting.. extends full computer search-and-seizure powers to cannabis enforcement.

This is interesting and the world models predicting the next frame is huge. I'm also wondering on how this would be applicable to neurological proceesses, thoughts etc.

If thoughts become quantifiable and measurable enough would similar models be able to predict the next one? Wondering what the impact of that would be.