For those who like to play with those trending tool …
A bad day to use python
I use virtualenv for every python code I download and test. Am I safer?
It is made to isolate Python packages required by each Python app from other versions that may conflict.
It doesn’t prevent access to the file system.
And litellm doesn’t need to be a direct dependency, any package using it with compromize the whole machine.
LLM’s and AI agent are a lot in fashion but they can be used as trojans later, even they look harmless for now.
You can still use them in rootless Docker or VirtualMachines if you block access to your filesystem.
But there are already a lot of data breaches and exploits due to agents managing emails/documents!
Neural networks are even more obscure than closed source.
Not even those who can access the code, can promise it won’t do any harm.
And now this is Axios which is compromized and Claude Code who leaked …
AI is making the things more and more crappy!
1 Like
An example of AI served at each meal.
Here is the presentation of ideas about to make “intelligent” elevator for blind people.
https://www.trikalaonline.gr/trikala/όραση-για-όλους-πρωτοποριακή-και-καιν/
Another stupid idea based on artificial intelligence under the guise of “social equality”.
Blind people are resourceful and adaptable.
They have never needed a digital assistance system that would only make them less independent.
Furthermore, they still need someone (whom they trust) to accompany them when they interact with public institutions.
This is yet another way to reduce the social connection and interaction that occurs when they ask for help.
Let’s isolate them by subjecting them to Artificial Intelligence, it is good for business.
Besides, it will be a system of extensive surveillance, as it will surely become the norm in the name of progress.
And soon, an AI assistant in the bathroom will tell you how to wipe your butt and check that you wash your hands.
Let’s start by allowing blind people to walk on sidewalks without potholes, free of trash, abusive coffee shops and restaurant deploying tables and chairs outside of the limits, and vehicles of all kinds parked on the sidewalks.
AI for blind people assistance isn’t a bad idea per se. However, the privacy aspect of it is really important. I think having a assistant-wearable that runs locally would probably be quite good. It is possible and there are some really nice use-cases for it.
On a side-note I am so sure that chat-gpt and the major llms are using people’s prompts to pry into their lives. Whether for less harmful things like directed Ads or more serious stuff. And it’s not just them, there is this really weird thing that many people notice. You talk next to your phone about some product you wanna buy, and magically you start getting ads for it. The whole system is compromised to be honest.
Maybe you didn’t read the full translation …
Specifically, the cutting-edge action aims to develop an elevator system that allows people with visual impairments to use the elevator autonomously, exclusively through voice commands.
Ok, voice commands but …
combines Artificial Intelligence, LiDAR sensors and voice guidance, transforming the elevator into a “smart assistant” that recognizes, guides and protects the user in real time. With a navigation accuracy exceeding 92% and a pilot application in real conditions, the project goes beyond the limits of an educational exercise. At the same time, it allows people with visual impairments to use the elevator autonomously through voice commands, recognizes the user’s voice, accepts commands for floor selection and provides audio guidance throughout the entire route, ensuring safety, independence and full accessibility in public buildings, educational institutions and high-traffic areas.
It already exists elevators with announcement of the current floor.
Blind people don’t need more than buttons in Braille.
Isn’t it “overkill” to add AI, LiDAR and even voice recognition+TTS (continous microphone) for something that is mastered already by the concerned people?
92% of accuracy? Only?
With a Braille button, this would be 100%!
This is not about to develop AI goggles for navigation, sadely.
It is just for the time that you spend in the elevator.
Another pretext for “we need AI everywhere” from people who imagine a pseudo-need for a category of disabled people of which they probably even never met a single one!
Like I often say, this is not because an idea looks feasible that it is a good idea nor that it has to be developed.
Next step: add rocket engines to the elevator to make it go faster!
Fun fact: the city where I live is nicknamed “έξυπνη πόλη” (eksipni poli, the intelligent city) because we have some infrastructure like 5G and optical fiber.
They even do jokes about this “pride”, like on this article of … today! https://www.trikalaola.gr/epistoli-idou-i-exypni-poli-trikala-tou-dimou-trikkaion/
But in a big part of it, it looks that third-world, a lot far from consideration for any disabled people.
The quotes are translation of excerpts of the article on the previous message.
As you can see, a smart elevator is what blind people miss the most for their autonomy in Trikala!
The only justification of the need of Artificial Intelligence in that case is the lack of Human Intelligence.
Put as much of AI you want in an elevator, let’s create a marvelous marketing speech about autonomy, accessibility, dignity, safety, inclusion and all the “progressist” bullsh*t to sell a product that no one need, even the “target”, and they hope to make it a “technology model that can be applied throughout Europe.”
If they want to make the life better for disabled people, they should educate ordinary people to respect them to make their life easier everywhere, not only inside an elevator of public buildings, which is infinitesimal part of their life.
Oh yes, I completely missed the article lol. Agreed, in this specific use case, AI isn’t needed. At most a speech-command system could be helpful. That article seems to be a classic case of over-engineering a solution to a non-existent problem. I was talking generally about AI-assistants for blind people.
There are open source ai models in Google, hugging face, kaggle, etc
not the latest and greatest, but they are testable, executable in your own server and the community could put their stamp of approval
And this malicious python code can be mitigated by running it on its own non root user, allowing only one network address:port
The issue is not so much about AI by itself, but about “Supply Chain Attack”, via dependencies of dependencies, you can’t always have a check on everything that is included, unless you use fixed version.
It is a matter of trust in packages systems.
In the case of Axios, it took about 3h before to detect and remove the trojan infected version …
Besides, having AI agents running on you machine is like having a spy/sleeping agent wanting for the correct prompt to do harm.
A universal pre-installed malware toolkit ready to use for the attackers!
AI, AI everywhere …
At $599, plus an annual subscription ranging from $70 to $156 (…)
Subscription … for the updates …
Let’s hope that they don’t produce “Blue Water of Death” when crashing