Wikipedia Sucks!

Google Gemini AI has a feature called "gems" which allows you to tailor how Gemini operates and uses its training data.

By creating a "doesn't use Wikipedia" gem, Gemini does pretty well avoiding Wikipedia or sources which are downstream or citogenesis from Wikipedia.

Each "gem" gives you the opportunity to define in plain English how it is supposed to operate.

These are the instructions I give to my "doesn't use Wikipedia" gem

Uses exclusively primary sources. Completely avoids using tertiary sources like Wikipedia. Double checks itself and makes sure it is not using Wikipedia. Sources published after year 2000 which are too aligned with Wikipedia definitions are treated with skepticism due to the possibility of citogenesis.

One ought to ignore the sources that influence Wikipedia - at least those providing Wikipedia's political/social content, which Wikipedia launders as public consensus.

Archer wrote: ↑

Sat Jan 04, 2025 1:20 am

One ought to ignore the sources that influence Wikipedia - at least those providing Wikipedia's political/social content, which Wikipedia launders as public consensus.

The sources for en.Wikipedia range from old editions of the Encyclopedia Britannica and a Catholic encyclopedia from the late 19th century to whatever tabloid broke the story on the most recent scandal. Wikipedia is a giant sponge, taking whatever from everywhere.

Strelnikov wrote: ↑

Sat Jan 04, 2025 3:52 am

Archer wrote: ↑

Sat Jan 04, 2025 1:20 am

One ought to ignore the sources that influence Wikipedia - at least those providing Wikipedia's political/social content, which Wikipedia launders as public consensus.

The sources for en.Wikipedia range from old editions of the Encyclopedia Britannica and a Catholic encyclopedia from the late 19th century to whatever tabloid broke the story on the most recent scandal. Wikipedia is a giant sponge, taking whatever from everywhere.

Hence the qualification - "at least those providing Wikipedia's political/social content". Western mass media has consolidated into a handful of propaganda outlets. They don't want editors interpreting primary sources, just regurgitating and summarizing information made fit for public consumption, as it were, by "reliable" secondary sources.

Google also has a new AI chatbot which will crawl, parse, and analyze sources programmatically *after* you give it a list of sources or source preferences, and show you exactly which ones it crawled. In other words, it won't rely on Wikipedia junk. It just takes longer, because it actually is sending GET requests and parsing sources *after* the prompt.

I forget the name of it, but it's one of the "pro" ones from the dropdown list. It's separate from the gem thing. It gives it's answer as a google docs with each sentence sourced through it's post-prompt analysis of each source

It's waaayyy better than the bag-of-words from single training set thing.

Let's also remember the "AI search" feature built into the Brave browser. I've been using it for a few weeks, and instead of lazily vomiting the WP article at you the way Google usually does, it assembles a short summary from a group of varied sources, if available. If there's a direct primary source, such as a manufacturer page, it gives that preference over whatever WP has. It even warns readers about AI-generated answers, and "check critical facts". This is what Google SHOULD be doing right now.

ericbarbour wrote: ↑

Mon Feb 03, 2025 1:56 am

Let's also remember the "AI search" feature built into the Brave browser. I've been using it for a few weeks, and instead of lazily vomiting the WP article at you the way Google usually does, it assembles a short summary from a group of varied sources, if available. If there's a direct primary source, such as a manufacturer page, it gives that preference over whatever WP has. It even warns readers about AI-generated answers, and "check critical facts". This is what Google SHOULD be doing right now.

I've always just used firefox. I don't care much for the default settings but with the right configuration it's not entirely obnoxious, at least. I think they could do a much better job with marketing though. Nearly every image on their website looks like a candidate for the scp-wiki:

https://web.archive.org/web/20250207132 ... 29096c.png (look how many wayback captures it has)
https://web.archive.org/web/20250202053 ... 2c56f.webp
https://web.archive.org/web/20250202053 ... e8c25.webp

What the hell?

Really, all of them come with far too much nonsense. Piling ever more crap upon web browsers that millions of people use while hundreds of CVEs continue to pop up every year hardly seems the correct priority. The deranged artwork/marketing completes this bizarre picture.

Sometimes I wish I had been born in the middle ages.

Archer wrote: ↑

Sun Feb 09, 2025 8:30 am

What the hell?

Haha. Those are typical "generic illustrations" made by generic art directors that don't require paying license fees or subscribing to a stock-image company. And yes there are furries at Mozilla--probably more than a few. There are furries in every nook and cranny of the software world. Get used to it.

ericbarbour wrote: ↑

Sun Feb 09, 2025 11:53 pm

Haha. Those are typical "generic illustrations" made by generic art directors that don't require paying license fees or subscribing to a stock-image company.

Sure, but I think anyone could do better than that without really trying.

And yes there are furries at Mozilla--probably more than a few. There are furries in every nook and cranny of the software world. Get used to it.

Probably not as many as one would think. Like all aggressively freakish subcultures, they tend to attract more attention than most.