People who are against #AI are against progress itself. Sure, it has its issues but that's because those who are unable to see the positives simply do not look for them for example, AI is the way that me as a #Blind person is able to get descriptions that sighted people wouldn't want to write themselves. It is the way that at some point, i might be able to drive, something which sighted people take for granted because they can just get a license and drive where they want, while I have to struggle to even get to the grocery store. There are so many benefits to it so I'm pro AI all the way. #Technology, #Progress, #Computers, #cars
reshared this
Matt Campbell
in reply to Nick's world • • •I'm sympathetic to this argument. I've taken the same position myself.
But we also need to look at the flip side. Progress, and especially individual convenience, at what cost? At the cost of continuing to raise the planet's temperature? At the cost of trampling on the rights of those whose words and pictures were used in the training data? I'm not sure.
Matt Campbell reshared this.
Talon
in reply to Matt Campbell • • •Matt Campbell reshared this.
Nick's world
in reply to Talon • • •Nick's world
in reply to Matt Campbell • • •Matt Campbell
in reply to Nick's world • • •I hear you, and I've made that argument myself. I'm just not sure if that gives us the right to take what we want by brute force, at the expense of other people and our world as a whole.
If one of us needs to use an LLM to get past an accessibility barrier that's standing in the way of employment, education, or some essential task, I don't have an argument against that. But using LLMs for convenience or entertainment seems harder to justify.
Nick's world
in reply to Matt Campbell • • •Nova🐧✨
in reply to Matt Campbell • • •i think it's in many peoples' best interest to make LLMs and diffusion models and all more efficient and less power hungry, it's not like crypto where proof of work made using up as much electricity and hardware as possible the more effective method... we now have NPUs that accelerate usage while remaining under 10W, and models that are combined with novel techniques can outpace newly trained models with negligible energy consumption.
Once the AI bubble pops and big companies like OpenAI inevitably fall under their inherently bad business models, the open source models and distributed compute for training will still exist and will get more and more efficient with real use cases that people care about.
reshared this
Matt Campbell and Mikołaj Hołysz reshared this.