As if AI itself is not bad enough

[cat-man (Mikael)]

yes if i find something abou it ill share it here. it says if you talk to chat gpt and others you can upload pictures to it. and to upload picture you need to be logged in and i dont want to logg in to those models the only model i logg in to is protons Lumo

[JoelA]

I’ve used AI a few times to convert documents that were an image file- (jpeg etc..) into doc files so i could edit them. Also hand written notes. It is pretty amazing- Much more accurate than OCR.

[GrahamLees]

and so the worrying influence of AI continues as demonstrated in this Cybernews article

[josephj222222]

@cat-man

Im curious how does an AI read a picture i haven’t done it myself but many AI has it so people can upload pictures to them ,im ok with that an AI can read text or numbers but how can they tell what is on an uploaded picture ?

I have flameshot installed on my notebook. It takes a screenshot, does OCR on it and puts results in my clipboard. Really yhandy for error essages, etc. that don’t let you copy them. Since that works, it may be similar to what these other programs are doing.

[easyt50 (Howard)]

@cat-man

Im curious how does an AI read a picture

I posted that question to Gemini. Here was the response.

An AI processes an image through a mathematical pipeline that converts pixels into numerical values, breaks them into patches, and aligns them with a language model.

• Pixel Conversion: An image enters the system as a massive grid of numbers representing color channels and intensity values (RGB).
• Patch Extraction & Vision Encoders: Modern multimodal models split the image into small grid patches (visual tokens) and pass them through a neural network (like a Vision Transformer) to extract low-level features like edges, gradients, and textures.
• Deep Feature Mapping: Hidden layers combine these patches progressively into higher-level shapes, patterns, and object concepts based on statistical patterns learned during training.
• Semantic Alignment: A projection layer maps these visual embeddings into the same semantic vector space as a text-based LLM, allowing the model to “talk” about what it sees

[cat-man (Mikael)]

wow sounds complicated

[nevj (Neville Jackson)]

@easyt50

multimodal models split the image into small grid patches (visual tokens)

The key feature there is that image patches are a 2D sequence, whereas language is a 1D sequence.. I guess that is why multimodal is required? Not sure?
The ‘weights’ would have to be 2D as I see it?
It is way more complicated than simple image processing tricks like filters.

[easyt50 (Howard)]

@nevj

mage patches are a 2D sequence

According to AI;

Image patches are flattened into a 1D sequence of tokens before entering the transformer, and multimodal architectures are required to bridge the distinct semantic spaces of vision and language using multi-dimensional weight tensors rather than traditional 2D spatial filters.

[nevj (Neville Jackson)]

@easyt50

multi-dimensional weight tensors

I would love to get on top of that.
It must have uses in image analysis, beyond extracting things from your screen image.

[easyt50 (Howard)]

I saw this today on Yahoo News;
“AI agent reportedly wipes entire season of Swedish TV show before premiere — as presenter says ‘I knew AI would be the death of me.’”

Having files deleted is bad enough whether by AI or some other mistake. One of my 1st thoughts was, Why they didn’t have a backup of the files? I did not read the entire news posting, but here is a link if you care to read about it.

AI agent reportedly wipes entire season of Swedish TV show before premiere —...

[nevj (Neville Jackson)]

@easyt50

, Why they didn’t have a backup

Agree.
I cant see how they could neglect something so important.?
Digital technology is not infallible, even if AI thinks so.

[cat-man (Mikael)]

And why in gods name did they let in an AI agent into their server and all the files and tv shows they store there

[nevj (Neville Jackson)]

@cat-man

let in an AI agent into their server

Well did they? It may have been an invader?
Maybe they were using AI to ‘enhance’ their show, and it slipped up?
Its poor security either way.

[cat-man (Mikael)]

yes indeed

[nevj (Neville Jackson)]

@cat-man

let in an AI agent into their server

We really have to learn how to manage this AI thing . It has 100x the brain power of a human , but no morals or commonsense. Letting it loose in a server is like asking someone to play the Devil’s Advocate, then letting them carry it out.
Of course it chose to wipe the disk clean … that is the easiest solution to any computer issue.

[cat-man (Mikael)]

if they did let an AI into their own server where every file and back up is they cant blame anyone but them selfs.

[easyt50 (Howard)]

@easyt50

AI agent reportedly wipes entire season of Swedish TV show

To err is human …

Read this update today, "the program director denied AI involvement, stating an internal investigation traced the deletion to human error at the production company coupled with a backup system failure. "

The edited version of the show was deleted but they had copies of the raw unedited shows that will take weeks to go back and re-edit.

[nevj (Neville Jackson)]

@easyt50

traced the deletion to human error

Still a major security issue.
In some ways I am glad AI was not involved … but it was a great wakeup signal

[Tom]

…“No doubt before too long, insurance companies will wise up to the risks and either hike up premiums or introduce exclusion clauses (or perhaps both).”…as if they need more reasons to increase our costs. All things digital can be hacked, lose their functionality due to atrophy, or be cannibalized. Have fun in your digital prison.

[pdecker]

I have what I think is a decent use for AI. I’ve done this a few times. When I drop to a terminal to update my computer, I list out the packages about to be updated. Then I asked Copilot to give a very short description of the packages being updated. In many cases, I already knew (or thought I did) what the packages were. In some cases, I didn’t.

It can be a fun way to educate yourself on your own computer.

Can you give a very short description of the packages that have updates on my computer?

cloud-init/noble-updates,noble-updates 26.2-0ubuntu1~24.04.1 all [upgradable from: 26.1-0ubuntu1~24.04.1]
ghostscript/noble-updates,noble-security 10.02.1~dfsg1-0ubuntu7.10 amd64 [upgradable from: 10.02.1~dfsg1-0ubuntu7.9]
libcares2/noble-updates,noble-security 1.27.0-1.0ubuntu1.1 amd64 [upgradable from: 1.27.0-1.0ubuntu1]
libgs-common/noble-updates,noble-updates,noble-security,noble-security 10.02.1~dfsg1-0ubuntu7.10 all [upgradable from: 10.02.1~dfsg1-0ubuntu7.9]
libgs10-common/noble-updates,noble-updates,noble-security,noble-security 10.02.1~dfsg1-0ubuntu7.10 all [upgradable from: 10.02.1~dfsg1-0ubuntu7.9]
libgs10/noble-updates,noble-security 10.02.1~dfsg1-0ubuntu7.10 amd64 [upgradable from: 10.02.1~dfsg1-0ubuntu7.9]
libnss3-tools/noble-updates,noble-security 2:3.98-1ubuntu0.3 amd64 [upgradable from: 2:3.98-1ubuntu0.2]
libnss3/noble-updates,noble-security 2:3.98-1ubuntu0.3 amd64 [upgradable from: 2:3.98-1ubuntu0.2]
libpoppler-cpp0t64/noble-updates,noble-security 24.02.0-1ubuntu9.11 amd64 [upgradable from: 24.02.0-1ubuntu9.10]
libpoppler-glib8t64/noble-updates,noble-security 24.02.0-1ubuntu9.11 amd64 [upgradable from: 24.02.0-1ubuntu9.10]
libpoppler134/noble-updates,noble-security 24.02.0-1ubuntu9.11 amd64 [upgradable from: 24.02.0-1ubuntu9.10]
poppler-utils/noble-updates,noble-security 24.02.0-1ubuntu9.11 amd64 [upgradable from: 24.02.0-1ubuntu9.10]
ubuntu-pro-client-l10n/noble-updates 38ubuntu0~24.04 amd64 [upgradable from: 37.2ubuntu~24.04.1]
ubuntu-pro-client/noble-updates 38ubuntu0~24.04 amd64 [upgradable from: 37.2ubuntu~24.04.1]

Sure. These are all pretty routine Ubuntu updates, mostly security and bug-fix releases:

• cloud-init - Initializes and configures cloud/virtual machine instances on first boot.

• ghostscript - Converts and processes PDF and PostScript documents.

• libcares2 - DNS lookup library used by network applications.

• libgs-common - Shared support files for Ghostscript.

• libgs10-common - Architecture-independent files for the Ghostscript runtime.

• libgs10 - Main Ghostscript library used by applications that render PDFs.

• libnss3-tools - Command-line tools for managing NSS certificate databases.

• libnss3 - Mozilla Network Security Services library (SSL/TLS, certificates, crypto).

• libpoppler-cpp0t64 - C++ library interface for reading and rendering PDFs.

• libpoppler-glib8t64 - GLib interface for the Poppler PDF rendering library.

• libpoppler134 - Core Poppler PDF rendering library.

• poppler-utils - PDF command-line tools such as pdfinfo, pdftotext, and pdfimages.

• ubuntu-pro-client-l10n - Language/translation files for Ubuntu Pro.

• ubuntu-pro-client - Manages Ubuntu Pro subscriptions and services such as ESM and Livepatch.

Bottom line: Most of what you’re seeing are security updates for PDF-processing components (Ghostscript and Poppler) plus security libraries (NSS). I’d consider these worthwhile to install, especially since PDF parsers are frequent targets for security fixes.