Some apps on here are offering AI/LLM for certain features such as entering an image description automatically.
If you are abled and sighted, please don't use these features.
If you are sighted and able to enter descriptions yourself, it would be better for the planet if you enter descriptions manually. Manual descriptions avoid anyone having to cause environmental damage.
@feditips building the AI model is what takes a lot of resources. Running the model is pretty minimal and often can be done totally locally on something as low-powered as a cell phone.
If you actually read the article I linked to, you would see it is about energy of actual usage (not training):
"And while we know training AI models is highly polluting, the emissions attributable to using AI have been a missing piece so far. That is, until now. "
"I just published a story on new research that calculated the real carbon footprint of using generative AI models. Generating one image takes as much energy as fully charging your smartphone"
@feditips That’s from scratch text-to-image generation, not captioning.
Eg, VoiceOver can run on iOS and basically describes what you see in your camera app in real time. Were it using as much energy as it takes to charge your phone, I imagine you’d notice.
I don’t necessarily know what tools what software is using to caption images, but most people can run some form of captioning locally without the CPU breaking a sweat.
@feditips Heads up, a lot of the work suggesting these outlandish water usage stats have since been debunked as overblown. Here's an article (https://arxiv.org/abs/2204.05149) that discusses some issues in methodology and goes further to point out that a significant portion of the ecosystem still has hardly been optimized. I understand criticisms about inhuman accessibility mechanisms and the potential for error/alienation, but I'm not sure we should be discouraging it on the basis of climate.
It's not just water usage, it's carbon as well. Anything that uses significant amounts of energy causes carbon emissions either directly or by using renewables that could have been powering something else that consumes energy.
"Generating one image takes as much energy as fully charging your smartphone."
@feditips One HUGE caveat to this is to note the difference between inference and training. The differences in resource usage are fairly massive, but training costs should be thought of as closer to an initial upfront cost rather than an ongoing expense of resources.
The article I linked to is specifically about the impact of using the models (not training but actual use):
"And while we know training AI models is highly polluting, the emissions attributable to using AI have been a missing piece so far. That is, until now. "
I don't know if that feature works any more? However, I'd imagine it causes extra energy costs of some kind.
By the way, just to repeat, I'm not saying any of this to people who depend on these features for accessibility. I'm only saying this to people who are able to enter a description manually.
Add comment