The unicorns, in my perspective, don't appear to have had any notable changes. It's interesting though, that we're assessing a language bot based on its ability to generate a drawing. After reviewing the blog post linked, I agree with the author's observation that there don't seem to be any significant alterations in the unicorn.
Indeed, there are numerous instances of developers experimenting with prompt engineering, discovering what methods work best.
However, I find it difficult to regard this as anything more than speculation for now.
This article is a little bit of a red hearing. OpenAI is not apple, in the sense that they are not great at building user facing products. They are great at building the world's best AI models. They've known this since the inception of the GPT models, the only way you could have accessed these models is via API.
Later last year, we saw the release of GPT's text-davinci-003, and in an attempt to showcase this new model to the research community, they launched ChatGPT.
I think what we are seeing now is that chatGPT is best when it is close to existing applications. For example what 14 year old is using the chatGPT app vs the Snapchat AI Chat which uses the API internally.
The recent drop in usage could likely be attributed to such preferential shifts, further compounded by the timing of school holidays.
I disagree with this slightly, modern computer chips follow a general purpose architecture not special purpose ones. The reason for this is building a computer chip is expensive and difficult to do. Similarly building any useful language model requires tons of compute power, and very smart ML researchers. Most of the smaller open source one's are just trained on GPT output.
By "cramming all of the web" on a model what is really going on is the hidden layers of that network are getting better at understanding language and logic. Imagine trying to teach a kid who doesn't know how to read to learn about a Science by only giving them science textbooks. Chances are they won't get very far.
Building little specialist model's don't really work either. It's like trying to train a parrot to do science, sure it can repeat some of the phrases that you give it, but at the end of the day it's not really making any new connections for you.
This doesn't work at scale. Stack overflow as a platform has been handling user generated input via moderators, voting, and testing. This is fine when there are only 26.8 million coders on the planet, most of which aren't posting on stack overflow regularly. With LLM's all of a sudden there is a huge influx of mediocre content on the platform that people can't handle. Inevitably this will erode trust in the platform. When someone posts a answer I assume they actually ran the code, and can verify the result. LLM's can spit out seemingly correct code that just doesn't work.
I know this crowd / silicon valley is accustom to these kinds of stories. But we should remember that this is a college kid, who saw an opportunity and took initiative to build a business. The comments here are WAY too critical. We should celebrate these kinds of stories, instead of nitpicking.
Additionally these 'A.I apps are just gold rush' comments are too dismissive about the potential. It may be that there are some low hanging fruit that will get picked first, but I personally think that the vast majority of opportunity requires some level innovation, creative thinking, and grit to utilize.
It combines GPT-4 with sources, and relevant images. It also has the ability to chat with historical figures.