Recently, I've discovered this: https://play.aidungeon.io/main/landing.
It's basically an AI assisted adventure story generator. I find it quite cool and exciting from a technical perspective, and I haven't felt this way about anything tech related for quite a while!
It got me thinking - there are probably so many opportunities and untapped potentials for AI and AI assisted projects out there, especially now due to the fact that AI is becoming more accessible to non-AI specialists (read - normal software engineers).
But it's also made me realize and ask the following question to myself - "What is out there in terms of accessible AI capabilities that I'm not aware of?" I wasn't aware of the possibility of producing something like in the link above until just recently. It's made me think - it would be really nice if there's a survey somewhere out there - a TLDR of AI and ML capabilities, from which one could dive deeper into a particular topic that interests one and perhaps implement it in a project.
I think it's possible these days, from my understanding, to not have deep statistical and mathematical knowledge of AI, but to be able to study a particular AI topic, such as a specific natural language processing or computer vision use case, and to implement it via libraries and having some level of understanding of the underlying concepts.
Does anyone know of such a survey or TLDR of AI capabilities?
The annual State of AI Report highlight many recent advancements.
https://www.stateof.ai/
When we talk about modern AI we are really talking about 1 capability: pattern recognition. There are three major ways it is applied:
Happy to discuss any part in more detail.
I work on machine learning for a while. TLDR don't try to develop custom AI because you need to spend too much money and time to develop even an MVP. Instead, develop something using GPT-3, AWS, or Google APIs.
I'm sure you could feed GPT3 a block of information and make it summarize it. Right now Open Cover Letter is generating cover letters with minimal training.
I don't know any but the message is basically anything you can get enough data on or find a clever way to generate the data either via simulation or via another context (maybe multimodal) you can basically do.