What I find super impressive now is image generation. I remember just 3-4 years ago the images generated by AI on news articles looked like LSD trip hallucinations. Now they can generate extremely plausible images of any scene you describe. It's absolutely wonderful.
RE being overtaken -- when ChatGPT-3 came out, I saw some startup doing "Automatic UI code writing with ChatGPT and design images", but after seeing OpenAI's multimodal GPT-4... it seems like that startup is no longer needed. I think this will be the case for hundreds of startups that are aiming to add 10% on top of CurrentGPT. Put differently, GPT-4 basically 10x'd GPT-3. So all of those startups doing GPT-3 + 10% are being left behind in the dust of GPT-4.
