LisaList2

Quote from: bmwcyclist on Yesterday at 06:37:48 PMClaude says:

Please don't do this. Anyone who wants to know what an LLM outputs for a prompt can use the LLM.

Are you not creating this tool yourself?

Quote from: Eschaton on Today at 12:51:51 AM

Quote from: bmwcyclist on Yesterday at 06:37:48 PMClaude says:

Please don't do this. Anyone who wants to know what an LLM outputs for a prompt can use the LLM.

I have the impression that the cost of using Claude to do something non-trivial is rising rapidly, so I appreciate someone sharing their experiments. Maybe my opinion will change, and I'm thinking that AI generated stuff might be best annotated as such so that the context is clear (and any that don't want to look at it have that option).

QuoteAre you not creating this tool yourself?

It certainly looks like Claude is doing some portion of the work and I suspect the rapid progress on this tool is very much a result of that.

How long would it take before someone achieved the same without Claude?

We don't have many people interested in creating new Lisa software/hardware, so my personal opinion (for now, at least) is that we should encourage anyone that wants to, whether they choose to use AI or not.

Quote from: sigma7 on Yesterday at 08:44:13 PMI think I've noticed that parts of some recent postings have a distinctly non-human style and now wondering if we should have a formal requirement to style such content so it is easy to recognize the portion that is AI output... comments/suggestions?

I think it is important to know if you are interacting with an AI, BOT or human. Which is why I clearly put is "Claude says". I think a policy of that type would be good for LisaList2 IMHO.

Quote from: Eschaton on Today at 12:51:51 AM

Quote from: bmwcyclist on Yesterday at 06:37:48 PMClaude says:

Please don't do this. Anyone who wants to know what an LLM outputs for a prompt can use the LLM.

If you follow the threads, the output was from vibecoding so the logical answer is from the AI.

.

Quote from: sigma7 on Today at 04:33:34 AMI have the impression that the cost of using Claude to do something non-trivial is rising rapidly, so I appreciate someone sharing their experiments. Maybe my opinion will change, and I'm thinking that AI generated stuff might be best annotated as such so that the context is clear (and any that don't want to look at it have that option).

I very much agree. I am about $300 into this so far and I doubt I am 1/2 done.

Quote from: sigma7 on Today at 04:33:34 AMWe don't have many people interested in creating new Lisa software/hardware, so my personal opinion (for now, at least) is that we should encourage anyone that wants to, whether they choose to use AI or not.

YES. Alex using Claude and providing examples for the Minesweeper app is what inspired my projects.

I personally think this could be a 'big break' for getting apps going for the LOS and a more active community. We will see.

I was worried about anti-AI when I made this post asking about posting vibecode here or on Github.

 https://lisalist2.com/index.php/topic,810.msg6045.html#msg6045

I DO think it is very important to be clear about what is AI and what is not.

I am not a developer, so I could not even begin any of the projects I am working on, and so far as I can tell, no one else was currently working on them.

All of it is built on the generous sharing of previous projects by others who are more knowledgeable and talented, as I always make clear.

I was given the advice (see thread link above) to use the tool to learn about the Lisa OS and development as I went, and that is exactly what I am doing.

Quote from: TorZidan on Yesterday at 06:45:00 PM

Quote from: bmwcyclist on Yesterday at 06:08:42 PMOne thing: the files downloaded by the LOS app are indeed available in the Workshop. That is how I update the application! Unless you mean you don't want to switch to the Workshop from LOS? I am hoping this will also be useful to folks who are not developers. I do have a workshop version; it's just clunky without the interface

Once I upload some source code file to the Lisa, the next logical step is to compile it and link it. This can be done only in Workshop (asfarasiknow). It does not make sense to boot into LOS, receive the file, then reboot into Workshop to compile it.

I see what you are saying, and that makes perfect sense.

For me, I need to go into LOS to test the code, and most of the time, the changes I want don't work.

So I write some problem descriptions, look for sources, request new code, and download it while I am still in the LOS, using a known-good version on my floppy.

So for me, there is no wasted reboot.

Early on, the code was failing during compile or link, but that rarely happens now.

I did have a version that was Workshop-only before I got far enough along to get it into the LOS. I will try to publish that also! Thank you for the feedback!

.

Quote from: Eschaton on Today at 12:51:51 AMAre you not creating this tool yourself?

I am not sure how to answer this. So far as I know, there is no big shiny button in Claude that I can press to make an app and then poof, it's done.

I have spent many, many hours working with the AI, debugging, finding sources, getting help from people here, and I doubt I am only 1/2 of the way to the end of my first "real" app.

I suspect people with coding experience might have been able to do this over lunch, but it has taken me all of my free time, even lunches at work, every day for a couple of weeks.

So I would not say Claude has done this alone, or, as far as I can tell, the AI could not have done this alone. I would also say I have put more time into this than many of my other projects. I certanly could have recapped a pile of Mac PSU and motherboards by now in less time.

That said, this has been one of the most inspiring things I have done in a long time. I have learned a great deal, and I can't express how the jolt of excitement hits me when a new feature works and I get to see something brand new in the LOS.

Quote from: bmwcyclist on Today at 01:18:40 PMFor me, I need to go into LOS to test the code, and most of the time, the changes I want don't work.

So I write some problem descriptions, look for sources, request new code, and download it while I am still in the LOS, using a known-good version on my floppy.

So for me, there is no wasted reboot.

Early on, the code was failing during compile or link, but that rarely happens now.

I did have a version that was Workshop-only before I got far enough along to get it into the LOS. I will try to publish that also! Thank you for the feedback!

Tim, I don't understand the development workflow you described above. It seems that you are using LOS to both receive your source code and then compile it ? How do you run the Pascal compiler and linker from LOS? please share more details about your development workflow.

Quote from: TorZidan on Today at 02:11:49 PMTim, I don't understand the development workflow you described above. It seems that you are using LOS to both receive your source code and then compile it ? How do you run the Pascal compiler and linker from LOS? please share more details about your development workflow.

I apologize for my poor description. I am new to all of this.


1. I will be in LOS to test a new build. I will find anything that does not work or needs to be modified while testing.

2. I will ask Claude to fix the issue or tell me what it needs me to research or find.

3. Claude gives me the new code to compile.

4. I open up a terminal on my modern PC.

5. I will insert a floppy (image via FloppyEMU), run a known-good previous build from the floppy, and use it to download the new code to the Lisa. If this issue is not related to send/recieves I don't need the floppy image copy, I just use the already running copy of the program to download the next beta version of the code.

6. I exit LOS and reboot into the Workshop.

7. I compile and link the new code, exit Workshop, and reboot into LOS to test it.

8. GoTo 1. 

Quote from: sigma7 on Yesterday at 08:44:13 PMI think I've noticed that parts of some recent postings have a distinctly non-human style and now wondering if we should have a formal requirement to style such content so it is easy to recognize the portion that is AI output... comments/suggestions?

Quote from: bmwcyclist on Today at 12:17:34 PMI DO think it is very important to be clear about what is AI and what is not.

I think that's a good idea. Nothing wrong with using it as an aid for your projects, but disclose exactly when and how you're using it. Everybody has different opinions on when it's appropriate to be used, and that way people can know exactly what they're getting into when looking at someone's project. For instance, I hold myself to the standard of never using it to write code for big projects that I really care about (LisaFPGA, ESProFile, ESFloppy, etc), but will absolutely use it to help with data analysis (it's been really handy for decoding and finding patterns in floppy disk logic analyzer traces that would've taken me ages to find), debugging particularly hard problems, and educational/fun/silly things like the Minesweeper app.