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.
“Artificial intelligence” companies are iterating quickly now. I’m definitely looking forward to the new memes based on GPT-4.
This is what Chat GPT 4 responds to this article.
I have had very good success with the model, but of course there are still shortcomings. However as an “assistant” it really helps both in doing busy work, and also helping learn new techniques.
It’s pretty obvious to me that AI including this one will continue to become more and more intelligent in terms of being able to “pass the bar” and other industry certifications. “Watson” winning at jeopardy was a big milestone and AI has come a long way since then. But even so, I think AI naysayers will continue to be dismissive over the format, despite the fact this was the format envisioned specifically to test AI in the Turing test.
So IMHO two things need to happen to reach the general AI milestone:
1) The AI needs to become dynamic, continuing to learn more on it’s own after the initial training.
2) The AI needs to demonstrate some initiative of it’s own beyond that of a knowledge oracle. In particular it would think of things to do or say without just responding to prompts.
1 – because human intelligence isn’t static (although in the previous AI article I posted a link to a man who could not form new neural connections).
2 – because humans aren’t mere query boxes, we’re thinking all the time.
GPT-4 is not a “quick iteration”. It’s been cooking for a year or two and I don’t expect GPT-5 anytime soon. It’s just a bunch of companies pushing everything they’ve been tinkering out right now to ride the hype wave.