Homeownership is still better than renting in an environment where landlords won't be punished in the market for neglecting their properties

Sadly, I think that's already happened in the US due to inflation in housing, healthcare, and education making talent so much more expensive here

Agreed. And it's one of the reasons I don't think AI is going to reduce demand for developers long term.

That was my impression when I started as well. The newer, code specialized models are better and I've even managed to get it to dash off some working ad hoc scripts that would've taken me a couple of hours in a few minutes.

That's funny, we had a similar "mishap" at my employer as well. I agree that it shouldn't be the only thing you use. It's better than "hello world" docs for getting started, but you have to branch out quickly to refine things. I mostly like it for documenting my stuff and writing help files.

The crazy thing is that they are similar algorithms applied to massive amounts of language data instead of more discrete data sets like other ML applications.

This is one of the reasons I think it will always have a place for people learning to code. The newer models even explain what code does line by line

One of the things that gives me hope these days is that the people with actual skill still aren't desperate enough to have to work for this unpopular administration

In some ways that's the worst part of it

My (semi-educated) guess is that it doesn't. Instead the human language model interprets the language of the request and then formats it to send to a specialized ML model trained on an optimized and tagged dataset of DNA strains.

Nowadays the LLM is a general front end interface that parses your prompt for the information needed and then programmatically sends requests to other more specialized ML systems that were trained on optimized datasets. In a lot of ways LLMs are a bunch of ML systems in a trench coat.

You're correct that LLMs are prediction devices. They do the same thing as other ML systems but with words and phrases from the training data instead of images. The problem is the training sets are so huge it's impossible to manage the "correct" and "incorrect" of the data.

In layman’s terms, an AI is trained by reviewing thousands of pictures of scans that are known to be positive or negative. When it reviews a new scan, it will tell you how well the new scan matches mathematically to the patterns it found in the positive and negative scans it trained on.

The conservative media apparatus always lashes out when the other side copies it’s methods because they know it would work.

To be fair, they are not really wrong about those things. But a certain point you can't make perfect the enemy of the good.

Progressives were too busy hating on him for being old, not punishing Israel, or putting Trump in jail.