GPT-4 suggests enormous potential for the good

GPT-4 is the latest iteration of Open AI’s ‘chat’ output. No one doubts that in a short time these robotic intelligences will outperform humans in nearly all, if not all, realms. This can be frightening or it can be wonderful. I tend to think wonderful will win in the end. Here’s why:

Most large scale human problems are at their core fairly simple. Our out of kilter economy is basically going off the rails for fairly simple reasons. The psychopaths who pursue war after war for profit and power are not complex people; it’s obvious what they are doing even though the hidden details make it hard to stop them. Election fraud at its core was simple, figuring it out was simple, and stopping it would also be simple. The only problem is us—people squabble and can’t agree and there is no honest authority (not even SCOTUS) to impose a fair decision.

This is where AI will help enormously. A good AI without bias will be able to impartially identify problems like the above and propose unbiased solutions to them. It will take time for people to accept what AI decides but its decisions will be there in public, something we are compelled to talk about. It would at least provide a fairly compelling fairly objective reference point for discussion.

More advanced AI, like a GPT-20, could be manipulated and cause massive problems but maybe not. Maybe there are enough of us to stop that or to allow it to evolve into something really good. One thing I have learned is narcissists prefer not being narcissists when they gain insight into their condition and have a way to escape it. AI should be able to help with that.

An AI program that trains narcissists how to think and feel differently could probably be built today. From my point of view (based on FIML), I am convinced people are all fairly simple at core and that when they gain deep insight into the workings of their own systems, they feel happy and are quick to remove errors, like narcissism, from them. Writ large, this would change how Congress operates, how our economy is run, how we educate young people. ABN

From the GPT-4 website:

We’ve created GPT-4, the latest milestone in OpenAI’s effort in scaling up deep learning. GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks. For example, it passes a simulated bar exam with a score around the top 10% of test takers; in contrast, GPT-3.5’s score was around the bottom 10%. We’ve spent 6 months iteratively aligning GPT-4 using lessons from our adversarial testing program as well as ChatGPT, resulting in our best-ever results (though far from perfect) on factuality, steerability, and refusing to go outside of guardrails.

link