It's hilarious to me that people are willing to accept that technology can come as far as producing an LLM of the caliber of ChatGPT, but they can't fathom that it will continue to iterate and get better. It may be "bad" at certain writing styles right now, but give it some time. Think about how far it's already come for us to even be having this conversation.
Did you not read the article? Their whole point is chatgpt isn’t good at this type of writing because open ai intentionally trained it to do very non threatening corporate friendly speech. While code-davinci-002 (which predates gpt3.5/4) is objectively weaker this limitation wasn’t baked into it and it demonstrates LLMs can certainly write jokes and edgy humor.
No doubt better AI exists and is in development, but this article sets off my FUD detector; if it was that secret they wouldn’t be showing it off to writers, and if it was that good they’d be using it more already.
code-davinci-002 is not a secret advanced model; it’s in the GPT-3 series. You can access similar models in OpenAI’s direct API. And, I would hazard a guess that “base4” is in the GPT-4 series.
I actually agree with the guy’s thesis, and I’ve observed the same (that the un-safety-ized versions of these programs are a lot more powerful at creating good output than are the free chat versions), but the basic factual errors bug me (as does him saying “they seem to want to murder humans. They talk about it a lot” with no further explanation.)
"regulators should force studios to be transparent about their use of AI. Consumers deserve to know which art is humanmade, just as they deserve to know which eggs are organic and cage-free."
This is a really good suggestion. Governments should force businesses to signpost the shit out of their use of AI systems. At least that way we have full transparency and can have choice over which businesses to support.
@LollerCorleone Imagine 10 years from now. With 10 years of trained data and algorithms, faster computers. Who knows what breakthroughs and innovations we don't know will be available. You worry now? People will build cars with AI or replace the AI unit with their own, even when it becomes illegal. And now imagine what a world might be in 50 years. Man, this gives me a headache and they don't even exist!
AI will continue to improve. It may be seriously flawed now, it could never reach personhood level, but it will still be enough to replace a lot of jobs in all sorts of service and intellectual fields.
The executives will continue to push for its use, because it's cheap and effective. Company owners and investors won't accept being denied use of it, or that their profits are touched.
Lets not fool ourselves into thinking magic new jobs will replace those, we are already set to lose technological jobs that previous advancements had allowed us to move to, and anything more is going to take only a tiny fraction of people. If we can't even count on social and intellectual work, the best that will be left to people is to undercut machinery by taking on sweatshop jobs for cheaper than it would take to automate, and even that is not limitless. But people will not just sit on the sidewalk and die either.
Either our whole political systems and economy will change or future will be a horrible form of techno neo feudalism, where class inequality is unsurpassable. But it won't change without a fight, the people who are powerful today won't just let it happen.
The sensible way to go would be to just implement UBI, but that would take some sense and empathy that is just completely lacking in most people who could make it happen.
Add comment