First of all, thanks @Christian_Schmitz for the video of Geoff Perlman’s speach at MBS conference
I beleive that Xojo.AI.Help was a GREAT idea. Congrats.
Train an AI with pure Xojo documentation and converting notes as questions to train was AWSOME !!!
For developers it will be a more than usefull tool.
So @Geoff_Perlman please give us the chance to start using it as soon as possible. Also a beta version would be extremely useful. Try to let it be available today if that is possible.
And please it will be also very kind if you could describe and share the tools used and the processes involved in doing the trainning I will be very interested to get that information.
We are not quite ready for external beta testing as we are internally aware of enough issues we need to resolve first but once we are ready we will be sure to make an announcement. It’s very exciting stuff. It still makes a fair number of mistakes though so we need to get to the point where mistakes are rare before beta testing can begin.
Hi Geoff, will this run locally on our own machines or in Xojo cloud or someplace else? Also, I assume we would be able to opt out of its use and by doing so it would not be aware of any of the code that we ourselves have written, correct?
Thanks @Geoff_Perlman .
I understand your point and I know that for a final version you need to be quite confident with the training and the results, but notice this two things:
Most of us, we are using the public not so good chatGPT Xojo trained AI assistant, so any specific trained AI would be probably better or at least could be another source of information.
Beta testers could help you to better teach the model with the results received to many prompts that good or not so good, at least by now, the AI returns.
I believe that understanding that it is a beta and needs more training, having a beta been used for many Xojo developers would be a win-win situation.
Thanks you answering so fast.
It runs in the cloud and does nothing without your explicit consent. Think of it like a Xojo expert who patiently awaits any question you might ask. If you don’t ask a question, it does nothing. If you want it to analyze code, like any expert it will be up to you to provide it with that code.
According to Geoff’s talk at the MBS conference it works by feeding off the documentation only. It does not use the forum and does not feed off anybody’s code.
Im new here and I was just going to ask for an AI deeply trained in XOJO code. I’m not much of a programer, I was never able to sit for long periods due sciatic nerve problems. However when I was younger I did some programming and access and several other object based programming languages with a similar interface a similar style of programming. I would like to see a highly optimized AI that can run on my desktop with a connection to the web that is only used as needed. Sort of a hybrid Desktop/ cloud approach.
The goal is for it to be free for all users. Whether or not be can achieve that goal remains to be seen. We are focused right now on using ChatGPT 3.5 because the cost is so little that it should make it possible for us to provide it for free but it has to do a good enough job using 3.5. We will see.
That shouldn’t be a criticism. If you incur long-term and significant costs as a result, then someone has to pay for it. And that should be the users first.
If a subscription for this cannot be avoided, then that’s a shame but understandable. Perhaps the LLM can at least enrich the online help search in this case.
I find the thought that people would be willing to pay for error-laden text prediction based on documentation that is missing information and is sometimes flat out wrong kind of offensive.
If you’re willing to pay for a solution, pay a human.
Quite frankly, this is not where I would prefer Xojo spend their resources. We don’t need them chasing the “LLM all the things” bubble, we need things like a practical way to share code between project types. We need a renewed focus on cross platform instead of multi platform. I know better than most that the efforts of one engineer doesn’t necessarily detract from the efforts of another, but Xojo has such a small team these days. It’s hard not to imagine that this endeavor is taking engineering time away from things that are actually useful.
This isn’t a “LLM all things” bubble. The impact of being able to access the documentation via an LLM that is trained upon it can’t be overstated. I’ve spent quite a bit of time using it internally in its current state and it’s pretty amazing. It’s like having an expert Xojo user sitting next to you waiting to answer your questions. The beginner can ask it beginner-level questions where they don’t even know Xojo terms but instead put things into their own words. The more advanced users can get quicker answers, get it to help them find bugs, review their code to provide suggestions and more.
I’m convinced that the way in which people access documentation today is going to take a backseat to doing it via an LLM. It is SO much more productive. The docs still need to be there not only to train the LLM but also because sometimes you want to review all there is to know about a particular class for example or you need the syntax for a specific class member, but for getting answers to most questions, doing it via an LLM is WAY more productive.
And for FWIW, this isn’t currently taking ANY time away from the engineering effort as I’m heading this project and working with an outside consultant to get it working.
*