For example, I hear some people say that it is great at code synthesis, but in my experience it goes completely off the rails if you ask it to write something that's even a little bit unlikely in the training corpus.
For example, I hear some people say that it is great at code synthesis, but in my experience it goes completely off the rails if you ask it to write something that's even a little bit unlikely in the training corpus.
* Coming up with design patterns given instructions on what I want to accomplish.
* Generating skeleton code.
* Code review up to a point (it's often wrong, but also give valid suggestions)
* Generating C# documentation comments for methods and classes.
* Generating in-code comments (i.e. //)
* API naming ideas (classes, methods)
* Giving me names of algorithms that I might never have heard of. For instance I wanted to partition some data evenly into buckets so I told me about the first fit decreasing algorithm, which was exactly what I needed for what I wanted to do.
I see it as an impressive toy at this point. Not to downplay the work they've done an it's certainly progress but it suffers from the same problems as all AI so far. Seems good mainly for spam or for automating writing another dozen of those horrible motivational books. It doesn't work for search/research, it's still faster and less troublesome to do that on your own. At least if you care about quality results/correct information.
1. Little programming tasks. I.e., write me a javascript function that extracts the tweetid from a tweet. Beats Googling it and then having to research through 15 years of Stackoverflow posts
2. Email upgrades. Sometimes I write something but it doesn't sound right, so I ask ChatGPT to rewrite it. Sometimes it works sometime it doesn't. It's a timesaver nonetheless.
With the appropriate prompt I can literally just toss the entire block of text into chat GPT and it will spit out a polished structured version of my original transcript, I've probably saved dozens of hours already.
It does an excellent job summarizing and restructuring existing data.
I have heard people using it for fiction generation, but the results that I have seen are highly unimpressive and tend to be extremely formulaic. Impressive for a LLM but not for a good storyteller.
But, it's a lot of fun the first couple of times. I recommend everybody to play with it. Have it plan a trip, write a short sci-fi story with your colleague in the lead. Ask it about how it can "think". It feels magical.
I open a tab in the morning and ask it things throughout the day, whenever I need to search for something.