OT: How AI is shite

Dunno what you wanted but:

At least two of those are selling you the motor, and another discussing how to DIY.

Not to disagree that searching hasn't got worse, in particular giving up after 3 pages of results and claiming your search term is nowhere else on the internet.

When looking to buy an obscure car part the other day, I was surprised to find DuckDuckGo (which uses Bing) found a plausible seller that StartPage (uses Google) didn't. That doesn't normally happen.

Theo

Indeed. Google is so heavily monetized that finding obscure items is very hard

replacement motor smart vertical blinds

And some of your search engines aren't AI.

Amazon, aliexpress and temu all deliberately add stuff that doesnt match. Ebay doesnt.

That's bullshit, google leaves them for dead.

This one throws up stuff a lot of others miss:

F****ng useless compared with google.

I have been trying to find the book that some of those 4 kids who were quite literally sold by their utter obscenity of a mother in the USA in 1948 when she decided to shack up with someione else who wasn't interested in her 4 kids.

Someone did name the book when the photo first showed up on facebook, but I can't remember the name of the book and nothing really searches facebook properly, even google, presumably because facebook doesnt let google crawl their site.

seachmixer is pretty f***ed, the first page is stuff for sale which includes 1948 all cars etc, no books when the search bar includes <4 children for sale USA 1948 book>

It does find the wikipedia article but none of the various newpaper articles at the time, but google does that on the first page and searchmixer still doesnt find that book.

Not sure that counts as AI - just bog standard web search. It has got worse as the "SEO" sharks and marketers have found ever better ways to game the search engines and fill their results with the crap they want you to see.

AI will allow them to make ever more convincing crap more easily - so it will probably get worse.

Perhaps we will be able to go back to buying the "Internet Yellow Pages" in book form again :-)

Yellow pages was always full of companies calling themselves Acrap, AAcrap, AAAcrap, AAAAcrap to get first listing in the section. Usually with no listed address and a national phone number to hide the fact that they were not a local company.

Its not AI its how things are referenced for searching, To them blinds are blinds. I should know because if I look for blind related products, meaning products designed for blind use, I still get blinds mixed in with the stuff I want. If I put unsighted, it does not know what to do. grin. Brian

I've been playing with kagi.com, which is a paid search engine (on the free trial). They do a better job of filtering some of the crap, eg they have a separate 'listicles' section for all the 'The N best XXX of 2025' and you can deprioritise or block particular sites (eg I don't want to see Quora because it's useless)

I tried my obscure car part and it found all the outlets my previous searches had plus one the others hadn't (an ebay listing that was out of stock) so for something run by a startup with a small set of users that's actually pretty respectable.

I tend to use heavily adblocked Startpage which is usable most of the time. I'll carry on playing with Kagi when I next have some difficult searches (you get 100 searches on the free trial, the paid plans are $5pm for 300 searches per month, or $10pm for unlimited)

Theo

Have a look at this:

Brian

I have notice how much slower Edge has become since they presumably added in all the AI nonsense.

It is effectively taking over all the memory and CPU power on my 4Gb Win10/32 PRO machine. Also hammers the disk drive even for quite a while after closing the entire browser.

Andrew

Are you sure it is locally implemented ?

How large would the LLM file be for such a thing, if it ran locally ?

What you could do, is you could ask Copilot, whether MSEdge runs queries locally :-)

So I asked Copilot that, and the answer was carefully crafted in waffle-words. But basically, no, the query is not running locally. MSEdge only "consults" the http: stuff it downloads, for building a page view, and according to Copilot, that's what it is doing today. Do I believe the answer ? Well, it depends on what the "training date" of Copilot is -- the info may not be up to date.

If "something is hammering your hard drive", this is excellent, because now you can use some forensic techniques.

Operation is CreateFile then Include Operation is ReadFile then Include Operation is WriteFile then Include

and then when you apply that to the trace, it will show the I/O activity and the process name doing it. Procmon has a neat "Jump To" function. If a trace entry mentions a filename for example, using Jump To will take you to the folder with the file in question.

If you save a trace for later as a .PML , you can save the original trace without filtering, or, you can save the trace with only the currently visible filtered items shown. I like to save whole traces, in case another filtering idea comes along later.

The only thing missing from Procmon, is info on which SVCHOST is which. You can run tasklist /svc and keep the output of that command, with your PML file, for later reference if opening the PML file again.

Paul