true, they don't have eardrums, yet...

the data was supposed to be to improve the model... not to sell me crap.

google are such scumbags and their share in the governing system of the steam client doesn't inspire confidence.

https://www.youtube.com/watch?v=jbw5FdbQkuk

AI Overview
To address concerns about an AI "listening in," it's crucial to understand that modern AI, including large language models (LLMs), is not a conscious entity that can intentionally eavesdrop or spy on conversations. Instead, it is a complex software system that relies entirely on vast datasets for its functions. The unsettling feeling that AI is listening is a result of advanced data collection, pattern recognition, and sophisticated algorithms that can make eerily accurate guesses.
How AI works and why it isn't "listening"
AI lacks sentience and consciousness. AI is not aware of itself, does not have emotions, and lacks personal intent. When a chatbot generates the phrase "I am hungry," it's producing a probable word sequence based on its training data, not reporting a physiological state.
It processes, not understands, information. Current AI models identify patterns and correlations in their training data. When given a prompt, they generate a response based on what they've learned from that data, rather than on personal experiences or an independent thought process.
The training process is disconnected from personal interactions. An AI's knowledge comes from its training dataset, which is then frozen in time. The AI does not learn from personal conversations and is not updated in real-time based on new, private interactions.
Personalization is based on your digital footprint. Many apps and services use AI to personalize experiences. This is done by analyzing your digital footprint, such as your search history, location data, and browsing habits, and cross-referencing it with patterns learned during training.
The privacy risks that feed the "listening" myth
The feeling that AI is listening often comes from legitimate privacy risks surrounding how companies collect and use data.
Ubiquitous data collection: Many apps and devices collect massive amounts of data—including voice commands, location information, and browsing history—to feed their AI models. You often consent to this collection by accepting a company's terms of service, though many users do so without reading them.
Overlapping data sources: A digital footprint is often pieced together from different sources. For instance, an AI might receive data from browsing habits through one app and location history through another to create a more complete—and often unnerving—profile of your likely interests.
Third-party vendors and data sharing: AI models sometimes use third-party datasets for training. This can lead to your data being used for purposes you did not consent to, or being scraped from public sources and included without your knowledge.
Data persistence: Once data is collected and embedded into an AI model's training data, it becomes nearly impossible to completely delete. It becomes part of the complex data structure that the AI relies on, though retraining models can reduce its influence.
How you can control your data
Several actions can be taken to gain more control over your privacy and the data AI systems use:
Adjust app permissions: Turn off unnecessary permissions for microphone access, location tracking, and camera access for apps that don't need them.
Review your privacy settings: Actively review and configure the privacy settings on your devices, apps, and browsers.
Use ad blockers and VPNs: To prevent some forms of cross-site tracking, use tools like ad blockers or virtual private networks (VPNs).
Limit data sharing: When possible, choose to opt out of data sharing with third parties, as many data protection laws now require this option

Or you could just make law making it illegal to harvest any data instead of opting out to 100 checkboxes you might miss

mood

it seems, according to gemini, I somehow "consented" to google using data collected from then AI when I created a google account before the AI even existed... I've now created a personal preference that's "supposed to" prevent my data from being used for marketing and/or targeted ads and/or external "personalized content"...

but this should really be an opt-in feature and I'm kinda disturbed a lot by googles liberties with my data.

and time will tell if it even works... if I see more targeted nonsense I'm outta there.

Originally posted by kingjames488:

the data was supposed to be to improve the model... not to sell me crap.

You know you can block entire domains, right?

Originally posted by kingjames488:

google are such scumbags and their share in the governing system of the steam client doesn't inspire confidence.

Wouldn't really call it governing so much as just being a base infrastructure for the client and super basic browsing ability. I'd rather have a custom coded client than use chromium.

Actually. 🤓☝️

There’s an AI that can find out what you’re typing based entirely on the sound of your keyboard.

There’s another AI model that can create an image of the room that a sound is recorded in. As in, you can play a sound to the AI, and based on the noise, the AI can recreate the room and stuff in it, that the sound was recorded in, based on the sound waves.

Both are eerily accurate.

Originally posted by Chaosolous:

Actually. 🤓☝️

There’s an AI that can find out what you’re typing based entirely on the sound of your keyboard.

There’s another AI model that can create an image of the room that a sound is recorded in. As in, you can play a sound to the AI, and based on the noise, the AI can recreate the room and stuff in it, that the sound was recorded in, based on the sound waves.

Both are eerily accurate.

Both tend to need full samples and heavily rely on the quality of the sound, something not many people invest in other than content creators, that dont tend to type on stream.

Originally posted by Mad Scientist:

Originally posted by kingjames488:

google are such scumbags and their share in the governing system of the steam client doesn't inspire confidence.

Wouldn't really call it governing so much as just being a base infrastructure for the client and super basic browsing ability. I'd rather have a custom coded client than use chromium.

same, this chromium-based client is buggy mess, I keep submitting bug reports like how in this thread the client randomly inserted one of my screenshots into this other users post
https://imgur.com/a/rkoYVzd

I send a bug report to steam and they're just like "we're aware of the issue, there's no fix".

like the bug where my PC actually soft-locks when trying to use chat in the steam overlay...

also fun fact, the AI has the steam forum indexed... it can even provide a summary of yout post history on steam.

though in being based on the chromium infrustructure it allows google to dictate steam policy such as which platforms the client can run on which then by extent of steam policies dictates who can access their games.

Originally posted by kingjames488:

like the bug where my PC actually soft-locks when trying to use chat in the steam overlay...

Had something similar but only the steam client itself, specifically when waking from sleep, the first time receiving or sending a message since wake would do that but no others. Was fixed in a version of the client after that.

Originally posted by kingjames488:

also fun fact, the AI has the steam forum indexed... it can even provide a summary of yout post history on steam.

It still can't hear my typing. I also like to block almost every google domain to avoid ads let alone all the other tracking stuff. Huge difference in cpu use when you block some of their scripts that don't seem directly related to video playback for places like YT which I like to avoid anyway.

Originally posted by Mad Scientist:

Originally posted by kingjames488:

like the bug where my PC actually soft-locks when trying to use chat in the steam overlay...

Had something similar but only the steam client itself, specifically when waking from sleep, the first time receiving or sending a message since wake would do that but no others. Was fixed in a version of the client after that.

Originally posted by kingjames488:

also fun fact, the AI has the steam forum indexed... it can even provide a summary of yout post history on steam.

It still can't hear my typing. I also like to block almost every google domain to avoid ads let alone all the other tracking stuff. Huge difference in cpu use when you block some of their scripts that don't seem directly related to video playback for places like YT which I like to avoid anyway.

the one you want to avoid is gstatic.

the only thing I've come across so far that actually requires it is gemini. it exists on every other google page but it's only purpose seems to be to track user data.