Is it all just fluff?
Over the last couple of months, the field of AI has seen insane amounts of progress. From OpenAI disrupting this field with chatGPT to Microsoft integrating it into their offerings and Google releasing Bard.
Let's get one thing out of the way, there is too much hype around this. More hype than is warranted for sure. And it's mainly driven by content creators wanting to ride this trending train.
But let's wait and think about what is new here.
AI is not new
Large language models, AI and deep learning aren't anything new. Big tech companies have been using these technologies for a couple of years now.
What is new however is generative AI. Which can absorb huge amounts of unstructured data and answer questions based on it.
What is hype-worthy then?
If you look at deep learning or machine learning models from a broad level. They do the following steps:
Analyze data
Create predictive models based on this data
Make themselves better over time
I would say the second and third points are still as they were a couple of years ago. The main difference is in the data. More specifically, the type of data which can now be analyzed.
Earlier, only structured data could be used for training these models. Even if you wanted to train them on data like Twitter tweets, which are unstructured. You would require huge amounts of effort to structure it first.
Now, we can also sort of run it on unstructured data. And this is huge since most of the data in the world is unstructured. Also, this was an area humans were better at than machines. Since so much data was unstructured, it was hard for machines to analyze it.
For example, people who trade in the stock market analyze data from news channels, the global market, technical indicators, and loads of other things throughout their day. They even extract body language data from how people are talking about the market online. And then they leverage the mental models they have created around this to make decisions about the stock.
This was earlier not possible by machines since they were only analyzing structured data. i.e. price of stocks, financial docs of companies, etc. But now, they should be able to analyze news articles, social media posts, and technical data for the past 50 years. Maybe it'll still take some time for them to analyze body language though.
The current state of things
AI is not really at a place right now where it can replace everything we can do as humans.
But what's a little concerning is that the level of advancement we were predicting was going to happen in 50 years, is now happening in just a couple of months.
Because of this, it has become really hard to imagine where this type of tech will be in 50 years. And it is this uncertainty that is causing all the chaos.