A critical look at whether the largest large language models justify the environmental and social costs when smaller AI models are already so useful