| I fully support the notion that OpenAI should be more "open" than "closed". I agree that OpenAI controlling one of the most massive and powerful LLMs right now is a huge risk. Especially for a company that's not particularly geographically distributed, as this puts the USA in a position of extreme leverage. I also understand that OpenAI may possibly be supplying unrestricted OpenAI ChatGPT models without any ethical or moral boundaries. For example, I wouldn't be surprised if the military had been training on ChatGPT for years already, on creating more effective ways of killing more people, faster, with a much lower cost. Granted, if you ask ChatGPT "What's the easiest way to kill the maximum amount of people with the minimum investment", ChatGPT will decline to answer. And I think that's good. What's not so good is the fact that these ethical boundaries are completely artificial and not built into the model. OpenAI can possibly activate or deactivate these boundaries at will. And it would not be surprising this is the case for governments or militaries. The great issue with this all, is that, while we can agree killing people is bad, there's other things that aren't so clear. For example: hacking. ChatGPT has actually declined to write a script that wasn't going to be used for malicious purposes involving the scanning of my own home network. And, like with everything, there's ways to break those boundaries, with so-called "jailbreaking" ChatGPT. Indeed, like many point out in the comments, a fully open-source ChatGPT may be desirable (it certainly is for me), but, with this, the likelihood of bad actors gaining control of it, disabling safety features (if there are any), and using it to do evil simply grows exponentially. In my opinion, the way forward is extreme regulation, Universal Basic Income, and other measures. Automation was supposed to allow humans to focus on more interesting work, and remove manual toil and back-breaking labour. That was the case for a while. Now automation is threatening to replace even highly skilled professionals like engineers, and/or make them become extensions of the "machine" by just giving it prompts (Prompt Engineering), or performing actions that AI models can't do like reading captchas. This is obviously extremely bad. Will open source solve this? No. |