Seems like an easy fix.

"The only ways to make the deletion stick are to disable Chrome's AI features through chrome://flags or enterprise policy tooling that home users do not generally have, or to uninstall Chrome entirely "

@atoponce,

> nano
> 4 GB

I'm not going to ask.

A gigantic rant but no useful, actionable information. Just like the last time. I'm going to ignore this.

Who's gonna tell him that Microsoft installs silently Copilot on Windows PCs?

find ~ \
  -type f \
  -name "weights.bin" 2> /dev/null | \
xargs stat -c '%s' | \
awk '{
  sum += $1
} END {
  printf "%.2f GiB\n", sum / (1024^3)
}'
2.67 GiB

Time for #degoogle and #diday !

#Google

I've no choice but to use Chrome for work on my self-supplied PC, but I'm currently trying:

- Delete weights.bin
- Create an empty file with the same name
- Set its file attributes to read-only, and system to hopefully make it impossible for Google's weights file to replace it when Chrome tries to redownload it.

Fingers crossed. If that doesn't work, I'll see if the hidden attribute works any better than the system attribute.

Another reason to use Vivaldi web browser.

It is the last mainstream web browser to not bundle an AI.

@Vivaldi

> At a billion-device scale the climate costs are insane.

this is just blind LLM outrage. Yes it's a waste of storage and network that nobody consented to. It's not guaranteed to make people use more CPU as they have to interact with whatever feature this is going to empower. But it's *only* 4GB.

Please write an article about the climate costs of internet ads. This is about a trillion times more wasteful and nobody consents to it. It wastes so much CPU and you could literally calculate the damage to batteries of mobile devices caused by this unwanted, unsolicited, uninvited compute.

Is this news? It's over a year old. IIRC, it's designed to reduce the amount of data sent off-machine. They were talking about space consumption in Canary release in Feb of 2025 (groups.google.com/a/chromium.o…)

Looking for all the models that have been downloaded for cache in browser, it's not just limited to one. chrome://on-device-internals/

You can use the Tools tab to 'Load Default' model, then type your prompt in the box below. It's slower than a cloud-based response.

Ask 'how to spell "incorriggable" and it takes 2 seconds to spit out: The correct spelling is incorrigible.

Takes 10 more seconds to explain each part of the word.

Then it crashes.

I'd rather handle spell checks on-box than send every word up to a cloud service.

But, don't ask it what the range of radio delay is between the earth and the moon. Highest Quality says 2.5 milliseconds, and crashes. Fastest inference says 22 seconds, and crashes. Not usable.

Feature flag is here: chrome://flags/#prompt-api-for-gemini-nano

When enabled, it powers demos like this: chrome.dev/web-ai-demos/prompt…

#AI #GeminiNano

erm.. fuck you Sundar Pichai.

#Google #Chrome #SundarPichai #Nano #AI #Enshitification #DataTheft #Privacy #EFF #EU

I self host my repositories not really by choice, but because I'âe seen Sourceforge, Berlios, Google Code and a few other rise and fall. The same will happen of any big, centralized forge. It doesn't solâe the archival problem. It just accumulates a lot of things for a while, and then loses it all at once.

I try to make sure my projects are backed up by software heritage. There also used to be "phonebooks" like freshmeat.net (no project hosting, just links to wherever projects are hosted)

It is a good article.

But as a distro developer I am quite sad that such developer conversations neglect the role and the purpose of distribution projects.

Distributions do in fact represent a middle ground between going self-hosted and over-relying on a bit tech third-party.

They are made to solve the dependencies issues, to help with discovery, to serve as archives and as places to ask for help and collaboration.

Come to us, we have cookies 😀

Thanks for sharing! It was a great read.

Still a bit sad that the author does not link it to any "political" ideas or forms of organizations. In some (plenty) of ways github (and previous forms he mentioned) was a product of a decade that is gone.

I really wish there was more work on content addressable repositories and distributed identifiers for logins, mostly to make it easier to move, back up, or have more than one point of failure for that.

One reason I've been playing with Radicle is to try out one person's idea of how to do that for issues/repos/files. There are some interesting ideas, but it needs more polish (in my opinion, obviously), but has the potential of reducing that.

That said, I do self-host my own forge but I know its a single point of failure and has almost zero discoverability. It sucks because I'd like folks to use it, not deal with the hundreds of spam registrations, and the general crap involved with the process of self-hosting as opposed to writing code.

Self-hosting definitely has a cost of time and effort. Sometimes its annoying.

Going with a hosted service also has a cost, sometimes it is when they try to make a profit (I use to really love Gitlab, and I donate to Sourcehut).

Was it because of the the whole archiving thing? I read the piece.

Maybe that will eventually not be a problem with self-hosting.

I could see a future where Forgejo or something similar is federated like mastodon and people can work together effectively and fork across instances and host backups of each other's repositories.

Where it would be easy to search for public repositories across the network.

And in the same way it's easy to back up Wikipedia, any entity with the resources could archive the whole thing.

There could be a collective who's purpose is to maintain such an archive.