LOL *NSFW* *The new Demote thread*

Speaking of Meta:-
https://www.bbc.co.uk/news/articles/cly74mpy8klo

 

I’ve been collecting our grocery shopping receipts for a couple of months, because I want to try and be better in what we spend. I know the kind of stuff we need to stop buying, but I want to see the numbers. So I’ve got about 15-20 receipts from various supermarkets that I want to transcribe into a data set.

I detest the “A.I.” bubble, but I thought to myself “Well, this kind of thing seems like it’d be ideal to get an LLM model to do. It worked really well when I tried the image recognition feature in Mealie, it perfectly replicated a recipe from a photo on my phone - maybe I could take a look at their code and see what sort of ‘prompt engineering’ they’re doing in the API request”. So I got Ollama and Open WebUI running on my homelab, took a look at the Mealie source code, and then started to compile my receipts.

I quickly realised that it just wasn’t going to work.

I want one line per item I purchased, so if I bought two of a thing I want two separate lines to show that. I want the full brand and product name. I also want any loyalty card discounts to be a separate column, not a separate line.

Sainsbury’s receipts show multiple quantities of an item on separate lines, but any discounts applied are on the next line as a negative value. Sainsbury’s receipts abbreviate the product names, to the point where both me and my other half really struggled to figure out what some of them actually were.

Tesco, on the other hand, shows the full brand and product name - great. But it condenses multiple product quantities down to one line with a combined price. Its clubcard discounts are represented differently to Sainsbury’s, it wasn’t always obvious which product was discounted and the discount value appeared in a different place.

And then there’s the odd Morrisons and Asda receipts here and there, who both do their receipts differently to everyone else.

I would have had to do a hell of a lot of ‘prompt engineering’ (another term I detest), tweaking, customising, and verification. I realised pretty quickly that it was going to be far… far quicker - and more accurate - to just do it manually.

So that’s exactly what I did.

Well… OK… I admit, I got halfway through the receipts before I had enough and sacked it off for another day, but that’s just my ADHD “need shiny things” goblin brain getting bored.

Anyway.

I did it manually and got exactly what I needed. I’m not even going to try and waste my time with an LLM. I need accurate information, not made up garbage spat out by a glorified bot that owes its existence to wholesale plagiarism and copyright violation.

 

I see, laughing spread for your arse!

 

I’m not sure we have that kind of relationship yet, you could at least try buying me dinner first…

 

I work in the voluntary sector, and someone asked me to make a poster for youth volunteering. When I asked for pics, they just wondered why I didn't just use AI image generation instead.
I sat them down and showed them what AI produces: mutants with 12 fingers and fused hands, ranks of marching children dressed in military uniform or (and I think this one was thrown in to avoid any suggestion of racism...) images of black children wearing rags and carrying guns.

After that, the person requesting the poster went away to try to find some real photos for me...

 

Sid Snot: "So this bus breaks down, and the driver gets out to try and fix it. The lady bus conductress gets out too, and says 'D'you want a screwdriver?' - and the driver replies 'Leave it out, I'm late enough already..."

 

While waiting to disembark, this morning, my son nudged me and gestured toward the driver in the car alongside us - casually sipping from a can of Stella.

 

I got a feeling that 2025 is gonna be a hell of a year...

 

To paraphrase the late Sean Lock: "It's a good driving beer".

 

https://www.bbc.co.uk/news/videos/cly9vl9jjkvo

Turn your sound down before watching.

 

It's weird, but understandable, that the sound is more like shattering than a solid impact

 

I guess an iron based one would have been different.

 

There would've been a lot fewer tiles left for a start!

 

I have the option of enabling Apple Intelligence on my iPad, it remains turned off.

 

I've just been reading that the latest update to Word for Mac has Copilot switched on by default.
And if you turn it off, you get an "enable Copilot?" popup every time you start a new page!

What part of FOAD do these dingleberries not understand?