I don’t usually share what I build. Most of it lives on Codeberg or Github as a glorified off-site backup. 3-2-1 and all that. If I nuke my drive, I can pull it back down. That’s the whole point for me.
A while back I mentioned Pimaps here, a few people asked to see it, so I posted it. Got dunked on for AI slop despite explicitly marking which parts were AI-assisted (documentation and review- flagged with AIP). Similar story with Cliparr. I get it - Lemmy hates AI in any capacity.
What does the community actually expect from devs when sharing a project? I’m not assuming everyone here is a professional dev - though I’d guess the overlap is significant - so I’m wondering where the bar sits.
Polish? Full handwritten codebase? Something else?
How much of the sausage making do you actually want to see?
Meanwhile…
Anyway:
I think you’re finally getting to the realization: doing things without considering the moral or ethical implications is inherently, uhhh, immoral and unethical. Do you do more good than bad with your smartphone? Does it use thousands of watts (hours) every time you use it? Do you replace your phone every year with the new model, maximizing your environmental impact with the manufacturing? What do you think makes using a smartphone just as bad as using an LLM?
Look, I get it: considering all of the consequences of your actions all of the time is tiring. I don’t like doing it either. Clearly companies don’t do it at all (at least not for external consequences). Look where that has got us. Apathy is not an option.
Local models are better in many ways, yes. The local user has control over how much power they use (for the query, not on training), other resource usage (water, electrical power sources), what model (and by extent the training data it uses), and exactly what documentation they get out (depending on the query/ies). The best option is still obviously to write the documentation yourself, as the subject-matter expert.
Hang on though; you’re glossing over a lot of salient context.
See: https://aussie.zone/post/37281785/25338365
That’s neither here nor there. Let’s talk turkey.
On an individual level I actually think it’s worse. A lot worse.
Smartphone (Ericsson LCA): 57 kg CO₂e (Carbon Dioxide Equivalent) total over 3-year lifetime. Manufacturing dominates. https://www.ericsson.com/en/reports-and-papers/research-papers/life-cycle-assessment-of-a-smartphone
Smartphone (Carbon Trust): ~80% of footprint is production, embedded before you even turn it on. https://www.carbontrust.com/news-and-insights/insights/circular-economy-and-net-zero-how-can-carbon-footprinting-reinvent-the-mobile-phone-market/
Meanwhile - LLM inference: 0.3–0.4 Wh per query (short), up to ~17 Wh for long reasoning runs. https://arxiv.org/html/2505.09598v1
Even at 500 AI-assisted coding sessions (heavy runs at that), I’m looking at 3–4 kg CO₂e. That’s 4–6% of a single smartphone’s manufacturing emissions.
So no, they’re not “just as bad.” Individual smartphone ownership is orders of magnitude worse environmentally than individual, occasional LLM use for a hobby project. And people typically replace their mobile phones every 2-3 years . Which means every ~3 yrs your phone adds ~60 kg CO₂e.
A good primer on the topic: https://blog.andymasley.com/p/a-cheat-sheet-for-conversations-about
You’ve ignored all of the other impacts of LLMs while comparing different lifecycles.
If you’re going to include the manufacturing of smartphones, you must also include the “manufacturing” of LLMs. That requires everything from the construction of datacentres and the manufacturing of computers which go into them to the less “real” copious scraping, training, and shutdown that is required.
Here’s some other numbers that I found for a mix of ChatGPT 3.5 and 4 (since the stats for newer models are hard to find):
That puts the total per-card emissions around 3280 kg CO2e and the total usage around 32 800 tonnes CO2e just for one model. Per your smartphone metrics, that’s about equivalent to 560K people’s smartphone use. If you use a smartphone that is more carefully constructed, you can get that up to 1.13M. So for the same atmospheric pollution as 0.5-1 million people buying a phone, we made one model run for 3 years. Now consider there are multiple models and they usually get replaced more often than once every 3 years.
Most people rely on a smartphone these days, very few people need an LLM. Phones are significantly less polluting than an LLM datacentre so you’d be better off using your phone.
Now for other concerns, including the rest of the environmental devastation, which smartphones and most other products do not have. Most of these are still unknown or hard to quantify because the tech industry has long since abandoned open development of new products. I’m just going to list the problems because no one wants to read anymore paragraphs from me:
The only real positive of the tech industry shenanigans is that prices have gone up, which is discouraging people from buying these significantly environmentally impactful products and instead extending the lifespan of existing hardware.
why are you comparing manufacturing to inference? compare it to training.