neither of the big cloud models have any fucking guardrails against generating spam. I'd venture to guess that 99% of spam is either gpt3.5 (which is better, ch

neither of the big cloud models have any fucking guardrails against generating spam. I'd venture to guess that 99% of spam is either gpt3.5 (which is better, cheaper and easier to use than any local model) or gpt4 with scrapped keys or funded by stolen credit cards. you have no evidence whatsoever that llama models are being used for that purpose. meanwhile, twitter is full of bots posting GPT refusals.