LisaList2

I have a generic question that seems abstract but important: in the Office System, is the Desktop Manager the only thing that can make a file appear somewhere in the desktop user interface, with an icon associated with a particular program? (And the only way it will do that is if you tear off a piece of stationery or if you duplicate a file.)

(I guess the answer is no: when you install a new tool, that also sets up a new icon for the tool and for its stationery if it uses stationery. But that doesn't happen while the Office System is running.)

Quote from: sigma7 on September 18, 2026, 02:37:10 PMI'm conflicted about the AI issue...

It's definitely a complicated problem, and I'm not a neutral party to it. Anyhow, some miscellaneous thoughts follow, or maybe a bit of an essay:

I think that Codeberg faces a technical problem that LisaList2 probably doesn't, which is that AI scrapers find large code repository websites (like Codeberg, GitLab, GitHub, etc.) to be valuable targets and therefore place an immense strain on their resources. LisaList2, for all of its virtues, is not so valuable So for Codeberg, banning AI bots is a necessity regardless of whether it reflects their principles. (Codeberg does have a policy of not wanting to host AI-generated code on principle, but that's a separate matter.)

GitHub meanwhile faces similar problems and also has scraping countermeasures in place. That said, GitHub is orientated toward using code stored there for AI training: it's one of the reasons it's such a useful asset to Microsoft, and if people use GitHub's Copilot coding service, those interactions get used for training as well. (Lots of coding agents use coding interactions for analytics and training.)

I believe that not all AI scraping bots are created equal. I haven't sought out any recent analyses of the scraping ecosystem, but I suspect some are generally respectful of things like robots.txt files and also use measures like rate limits and scheduling to avoid overwhelming websites' resources. Others are quite the opposite and simply don't care about those things; indeed they make considerable efforts to circumvent blocking measures. Sidebar:

One of the most striking developments of recent years has been "smart" appliances that have a side job: using peoples' home internet connections to do web scraping as part of massive coordinated swarms. The term of art for this is "residential proxy network", and what this does is make it extremely difficult to block scrapers, since they now come from virtually anywhere --- and of course that's the point, to defeat blocking. Even faithful users of this website might have a new smart TV in their living room hard at work bothering SourceForge or something. This at last is why you (@sigma7) had such trouble accessing Codeberg: in light of this onslaught from everywhere, one of the last things that smaller-fry websites can do is require all visitors' web browsers to solve a computationally expensive maths problem before they can gain unfettered access to the website. Anubis is a popular free software package that websites use for this, and folks might recognise the cartoon mascot on the webpage from earlier encounters. Anyhow, if you browse the web with an old computer or with Javascript disabled, you can't easily solve the maths problem and so (fast or full) access is unavailable.

If this mechanism sounds a bit familiar, the maths challenge is very similar to the proof-of-work mechanisms inside a lot of cryptocurrencies, though that's the only thing they have in common so far as I'm aware. Signs are though that scrapers are getting more patient about solving maths problems, and of course smart TVs are only getting "smarter" with new hardware generations. Website operators report having to increase the difficulty of the maths problems that Anubis dishes out, which has the unfortunate side effect of being more conspicuous and annoying to legitimate human users.

So that means that you could make a distinction between AI services based on their respect for Internet etiquette --- not to mention other factors like their environmental record, their stance on sourcing training data, their attitudes toward safety and social impact, etc. At least, you could if you had enough of that information to hand. Often you don't. I wish it were different!

As for the benefit of examples of new Lisa programs:

At this stage, from a capabilities standpoint, I actually don't think there's much more for AI systems to learn about Lisa programming. As general-purpose coders they are now very good on their own, and I suspect they are more than able to grasp the details of Lisa Pascal/Clascal and other details of Lisa programming from documentation available online. Furthermore, as Alex has described elsewhere, any information the AI hasn't already seen or memorised, you can supply as input on your own (e.g. give it access to the Office System source code and to the ToolKit manuals). Some things are missing from the documentation that we've had to find out as a community (like my old post), but you can supply that too, and anyway if a casual-player rando like me can make a discovery, then a modern AI system can easily do the same if you give them the right encouragement and facilities (like emulators). And more and more people (and AIs) are learning how to do just that. To some, all this may be a controversial assessment, but I think it's an accurate one. I put it forward as a technical judgement, not a value judgement.

So, technically, I don't really think it matters very much if we try to share or conceal new Lisa source code from AIs. If we make the code available, they'll read it, but if they don't, they'll figure it out, perhaps at the expense of their users needing to pay the token costs of the AI's wheel-reinventing experiments. Arguably AI service providers would be happy enough with that 

So what's left? Disclaimer: personal opinion follows!

To me, I like the old computers I collect (my Lisas, the PERQ, the Tektronix vector-display machines, the transputer setup, etc.) because they feel like windows into a different universe where a different set of ideas took off to the ones we know today. When I use or write code for these computers, I'm often trying to imagine being in that universe and seeing what it's like. I collected a few NeXT machines in the late '90s with that same "what if" speculative enthusiasm. Then OS X took off and NeXTs became a lot less interesting to me. We found out the answer to "what if" --- it's this universe where we are right now!

For now, AI coding doesn't trade so much in "what if". If you have a program idea in mind for the Lisa, it will just go straight ahead and make it, overcoming any obstacle or point of incongruence between your idea and the underlying libraries and tooling. But those obstacles are traces of the realities, preferences, and philosophies that made the Lisa what it was when it was introduced. The people who created the Office System imagined a computer that worked in a certain way, and the ecosystem that they started would have evolved with those visions as a major influence. (Of course those visions were "wrong" --- that's at the root of why the Lisa failed. But what if?)

If I make a Lisa program, I want to run into those frictions that come from 40 years of divergent history and not being on the same wavelength. I want to try to get into its designers' heads. I want to have those moments of "so that's why it's this way". I think this kind of understanding is really helpful to trying to make something that would be a convincing "what if" Lisa program from, say, 1985. Could I do this kind of exploration by using AI to code? I actually think I could! But I'd have to be pretty careful to rein in the AI and keep it patient and "true" to the feel of the experience, and to allow me to discover along with it. AI coding agents are not made for patience --- they are built to go relentlessly and just get it done, no matter what "it" is. So you need to do work if you want to ride along with them on the scenic route.

Of course, that's my own preference and my own kind of enjoyment: it's not for everyone! And it won't take long to count the number of Lisa programs I've made since the dawn of capable coding agents: just one, my LisaVox demo, which I did by hand for the fun of it. And that was January. Coding agents have got a whole lot better in the eight months since then!

I'm totally in favour of people chasing their own pleasure and visions for Lisa programs and capabilities. AI systems have clearly revolutionised this: we went from maybe one or two Office System programs per decade to, like, ten in the past few months. People are dusting off plans and hatching schemes for more, and AI is making it possible. That's great! The only thing I'll say is that these new plans are inevitably going to evolve the ecosystem. @bmwcyclist's new utilities are great examples of this: they're setting up ways that people manage files on Lisas, and at some point they may become a big part of the "standard way to do it". Are these approaches the standards we want to wind up with? What are their pros and cons? I'm not hinting either way; I really have no opinion at the moment. What I would say though out of naked personal interest is that I hope we don't evolve in a way that makes it more difficult to achieve my "what if" fantasies someday. I have a few ideas for how that might happen (e.g. new convenience and interoperability libraries that start to subsume the original ones offered by the OS/ToolKit/Office System), but I've gone on long enough, I think! If anyone has read this far, I hope it's been interesting...

Quote from: berskyboy on September 18, 2026, 11:50:31 PMI got a new DMM let's try this again

Excellent; to see how it behaves compared to the other one, try these measurements again using the diode test mode:

Diodes	Fwd, Rev	Description
{BA, BB}	0.13, inf	CR20a, CR20b
{DA, DB}	0.30, 0.37	CR25, CR23
.
Transistors	Red-Blk	Description
{1-2, 4-5}	1.25	Q4, Q3 c-b
{2-1, 5-4}	0.62	Q4, Q3 b-c
{1-3, 4-6}	0.70	Q4, Q3 c-e
{3-1, 6-4}	1.89	Q4, Q3 e-c
{2-3. 5-6}	0.64	Q4, Q3 b-e
{3-2, 6-5}	inf	Q4, Q3 e-b

And for each of Q1 and Q2 now out of circuit, also measure

Red to Black
Lead 1 to Lead 2
Lead 2 to Lead 1

QuoteQ1
Lead 2 to case = erratic

"erratic" implies a bad probe connection. Perhaps there is another explanation, but first try again and see if pressing the probes more firmly against the transistor makes the reading stable.

Also, avoid touching both points of contact while taking the measurement. eg. don't 'clamp' each of the probes against the transistor leads using thumb and finger touching the metal parts. You could squeeze one of the probes against a lead, but the other should not have a conductive path to you, as the current path through your skin can affect some readings. If you can avoid touching both, then thats the best option.

This is true. I got my download from Sunder.Net and yes, it is circa 2022. This is mostly due to the fact that I do not have the technical skills or the tools to compile it myself.

As it is, I am having more problems with Lisa Workspace throwing errors when trying to compile a simple program, like ';' is expected and it clearly shows a semicolon ending the line. I am chalking this up to my knowledge is not up to par yet.

I got a new DMM let's try this again (I can try taking more measurements tomorrow). 

Q1
Red to Black. (diode test, 200 ohm test)
Lead 1 to case = 0L,0L
case to Lead 1 = 0.5, 0L
Lead 2 to case = erratic
case to Lead 2 = 0L

Q1
Red to Black. (diode test, 200 ohm test)
Lead 1 to case = 0L, 0L
case to Lead 1 = 0L, 0L
Lead 2 to case = 0L,0L
case to Lead 2 = 0L,0L

Quote from: sigma7 on September 18, 2026, 07:25:16 PMIf you're having trouble getting consistent results, that seems like a good option. Then you've got two to compare if something looks odd.

As it is, unreliable measurements aren't very useful and it is hard to know how to proceed with troubleshooting. If you can figure out how to make the measurements repeatable/reliable say 10 times in a row, then post what you get, and explain what you figured out that you needed to do to get reliable results. (Without an explanation of what was wrong, future measurements remain suspect.)

ok sounds good.  Do you recommend a certain brand?

Quote from: berskyboy on September 18, 2026, 07:11:38 PMI can try and buy another DMM, this one might be cheap?

If you're having trouble getting consistent results, that seems like a good option. Then you've got two to compare if something looks odd.

As it is, unreliable measurements aren't very useful and it is hard to know how to proceed with troubleshooting. If you can figure out how to make the measurements repeatable/reliable say 10 times in a row, then post what you get, and explain what you figured out that you needed to do to get reliable results. (Without an explanation of what was wrong, future measurements remain suspect.)

Q2

Red to Black. (diode test, 200 ohm test)
Lead 1 to case = 5 beep ,1.2
case to Lead 1 = 6 beep, 1.6
Lead 2 to case = 3 beep ,0.7
case to Lead 2 = 3 beep,1.5

I can try and buy another DMM, this one might be cheap?

Q1

Red to Black. (diode test, 200 ohm test)
Lead 1 to case = inf beep, inf
case to Lead 1 = 2 beep, 0.4
Lead 2 to case = 3 beep,1.2
case to Lead 2 = 2beep, 0.5