Scientific American has a pretty good podcast episode about AI
It’s at the very least a good discussion
1 Like
One important aspect, at least in IntelliJ, which I was referring to, is that initially, you need to essentially pose your question in a comment.
Suppose you have an ArrayList. No AI knows right away that you want to sort it, nor how you want to sort it. If you now write a command like:
// Sort this Array, case-sensitive, respecting German Umlauts, descending, etc.
you will receive your suggestions (usually with a slight delay).
Over time, it learns more and more about your code’s intended purpose, and the suggestions become more proactive. I was very impressed when I began writing my markdown documentation (user manual) for the project. CoPilot essentially wrote 80% of the content…
So, it is somewhat “intelligent” and not entirely dumb, but not intelligent enough to suggest that my program is nonsensical or that many solutions already exist. It won’t recommend using other algorithms, but it is surprisingly effective in speeding up repetitive tasks. Eventually, you’ll get used to asking questions within the IDE instead of consulting Google, which could be both a blessing and a curse of AI (leading to laziness and less critical thinking).
Taking your time is well-intentioned advice. When I first tried it in the middle of last year, I thought it was nonsense and wondered who would need something like that. However, after a few weeks, I experienced several “wow” moments, and now I appreciate the help regularly.
This is precisely what I believe will become a problem for Xojo sooner or later. For AI companies, Xojo will remain insignificant. External plugin developers might be able to create plugins, but their usefulness will be limited, and users will desire full integration into an IDE since anything else is rather impractical.
Good discussion, but Napalm is nothing special for instance and quite easy to get the recipe. You will find tons of it via search engines
And on stackoverflow, reddit etc. you find as well a lot of non-sense or outdated stuff. But it will be interesting to see, how AI will learn if you are improving suggestions. But then again: will that extra knowledge only be available to AI customers? Will companies “exchange” their knowledge? (I doubt it).
What I will definitely miss are the discussions why and which solution might be better and explaining the reasons etc.
And once only the AI knows “why” it wrote the code it did the way it did then … ?
If all the “knowledge” is in the black box we’re just observers looking at the darkness from the outside
That was just one example of how it can, or could, be used in nefarious ways
A relatively simple one but still just an example
If its on the internet ChapGPT etc can use it as source
NOW whether ChatGPT can do knowledge synthesis, extrapolation and other tasks remains, to me, to be seen
But I’m fairly confident people WILL push it in all sorts of ways
Without seriously considering whether they SHOULD do so
I’ll again refer to the BS course and their discussion about the fallacies that big data, which LLM is, can be very very very wrong
Lots of data doesnt mean its right
The lecture clips from this week starting with this video is a great watch (all of week 5)
Yes, but we all might have copied as a starting point code from an online search, if it has been posted legally on a blog is not always known to the reader etc. But yeah,
I’m not downplaying the risks and the dangers. But I think there’s also a lot of drama created for something that “scares” a lot of people because it’s so intriguing, but it’s still very dumb at heart. However, it is so dull that it actually occasionally hallucinates. AI seems to have taken reddit as a model here 
As that podcast notes we, as humans, do tend to trust it as it writes in a very authoritative style.
And thats a downfall
What or who does that remind me of? AI or NI?
And thats a downfall
Thanks for sharing your experience.
GitHub is a 30 day free trial and $100/year after that. My first couple hours with it I’d say it improved the quality of auto-complete a bit, but not enough to spend $100 on. However I think a few weeks at least is required to truly evaluate it and, as you said, I will get used to prompting it with comments. And it will improve with experience.
The solution I’m working on isn’t quite new, but just a couple of days old.
One problem I have encountered, which is also in fairness true of the built-in “intellisense” is that it sometimes inserts completions at other than the cursor position, but rather, at the start of the line. For someone who likes K&R bracing and tends to indent arguments, this is really irritating. But I think it’s probably a latency issue where it gets out of sync and can’t keep up. Usually I am busy tabbing over to indent something and then it takes the tabbing as acceptance of its suggestion and then it goes where the cursor was a couple of seconds ago.
All solvable problems, of course. These things will get better, both with new releases, and with more training.
That works very fine in IntelliJ. One would assume that MS is doing even better on their own products 
Yes, I encourage everyone to use it at least for the whole free month. At least on IntelliJ it doesn’t hurt performance wise etc. I only switch it off when going through tutorials, where I don’t want to get any solutions proposed
Cant you just hear in your head
If all your friends jumped off a bridge would you do it too ?
More shiny baubles for Xojo to chase might not be the best thing but …
1 Like
it is very likely but sadly irrelevant what they will or won’t do.
1 Like
Xojo is a death valley, a no news area. All Xojo News are: pure marketing. And nothing real new. Only promises. And promises may sound better authored with some weird AI
they are on it: Openai API (aka ChatGPT) - #12 by Nicholas_Pearce-Boby - General - Xojo Programming Forum 
Like I said: weird. Totally weird
All great innovation in Xojo comes from a superior NI. Focusing on AI, the good people may be mislead in the effort.
1 Like
Well to be fair that is one of their users that appears to be on it
Not them
Well I meant it is the max you can do with Xojo. Plugin for chatGPT. Not related to coding assistance. Sorry for the confusion.
Ah
I wouldn’t be surprised if once there is a plugin Xojo tries to do something with that to help you with your code
I wouldn’t be surprised if they pursue “something”
What sort of something remains to be seen since its patently obvious ChatGPT & there LLM’s do not have any decent understanding of Xojo code. So if you ask it for Xojo code for a particular problem you often get code that wont compile at all and has made up methods adopted from other languages that are sort of similar.
Will Xojo spend time to teach one of these systems about Xojo code ? lord knows but I expect not.
Will users ? doubtful. And if they do will they train it with “good code” ?
So I expect they’ll do something
How useful it will be … myeh… I’m skeptical at best