| >I'm not aware of any models with full training data available, let alone available with full licensing. Check out the OLMo family from the Allen Institute for AI: https://allenai.org/blog/olmo3 Also worth looking at EleutherAI, LLM360, SmolLM, Apertus. Apterus even respects opt-out notices in their collection of the dataset they distribute. KL3M from Kelvin Legal is even trained on a 100% copyright-free legal/governmental corpus. >23% of Ireland's power is going to data centers This isn't even a rounding error in terms of global energy usage. Besides, renewable energy is cheaper than non-renewable energy now. Energy demand itself directly drives the deployment of renewable energy. Needing more electricity in a capitalist country literally causes the energy mix to become more green over time. >And I'm also not aware of anyone working on these models trying to reduce the social damage they are going to do. Is it fair to account for the social damage without accounting for the social good, like the unprecedented democratization not just of mere information, but knowledge and understanding itself, around the world? Think about how many deaths will be avoided, how many billions or trillions of years of human life will be gained over time, all of the people that gain access to a personalized, individually tailored tutor for any subject on earth, all of the medical and legal advances, etc. |
Perhaps true, but meaningless except as a vanity metric if fossil fuel usage rises as well.