I am only seeing a steady and dangerous rise of AI in the open-source community.
Not a good sign if you ask me.
We are heading towards a future Uncle Theodore warned us about.
They're all just rehashing GPT in smaller models, pre-trained on different things to different degrees of optimization. The rest is just making UIs for the models, porting them for use in other languages/frameworks, ease of use stuff.
The bleeding edge research in AI is being done in universities or in small R&D projects focused on specific technologies that could be a part of a real AI system, or advancing brain scanning technologies so we can reverse engineer how biological neural networks operate.
So far, nothing big has come out of any of them.
Like 15 years ago there was all this hype about the work done on memristors at HP, people in the AI field thought it could open the door to silicon that acted more like biological neurons, both storing and processing data in the same integrated circuit. But so far, hasn't gone anywhere.
IBM's research with DeepBlue was hyped up, leading to versions of it beating a chess grandmaster, winning on the game show Jeopardy. This later lead to IBM developing Watson which had the ability to read large amounts of unstructured medical research data and "understand" it to the point where it could make accurate medical diagnoses and complex medical therapy plans for cancer patients (cancer treatment plans have to be highly individualized to the patient).
Its ability to consume regular documents and "understand" them was supposed to be a big breakthrough in ML, and it was a big jump forward. Essentially it was the GPT of 2011-2012. Then you basically didn't hear about it anymore. It's still around, still in use, but the hype is dead and the limits of the technology hit a plateau.
You had the Human Brain Project, which after like 7 years and around a billion Euros spent across 22-23 projects, has failed to generate a single noteworthy advancement.
Human-like AI is not easy and people playing around with NLP, vector databases, and RNNs aren't going to stumble onto it. They'll make a lot of automation that will replace a lot of jobs, but that's not even in the same ballpark as needing to worry about some Skynet scenario.