The Sinclair QL Forum

Derek_Stewart wrote: Sat Dec 28, 2024 7:48 pm Hi,

I agree with all the above points, in fact Qview same issues are stated in the Minerva Manual, in the Hatemail section:

Screenshot_20241228_194444_ReadEra.jpg


same issues 30 years ago

dilwyn wrote: Fri Dec 27, 2024 10:17 am ...

One reason people ask instead of searching here on the Forum is that the search (even the advanced search) is so hopeless at returning results relevant to what you search for. Even using search engines, I wish they had an "exact match" option so that the 99% of irrelevant results were eliminated, even if it often resulted in zero search results. Simple example: Try forcing Gongle to search for sinclair ql with a search such as "sinclair"+"ql" and 99% of what is returned has more to do with SQL or the Brother QL label printers than what you searched for. Despite putting words in quotes and using operators to try to specify exactly what you're looking for. Perhaps this sort of thing puts people off searching for themselves. I hate "fuzzy searches". And it's often worse with AI systems because they try to be too clever sometimes.

...

I think part of it is that more recently Google has added AI which, when you ask, tries to summarize to give you an answer. Unfortunately you can't always trust the answer (*** example below), but for some, if they like that feature and accept it, it's easier than sifting through search results (I find my students doing this more and more now) I would think maybe eventually as search changes with AI expected as an integral part of it, this could also proliferate to forums like this.


(*** So, I asked Google what the difference between Andrew Tannenbaum's 1st and 2nd edition was and Google AI said the second included source code in the text -- which I didn't quite believe. Then I asked if the 1st edition included source code and Google AI said yes...kind of a contradiction. I think it got confused because the 2nd edition, printed almost 10 years later, came with a CD of source.)

bwinkel67 wrote: Sat Dec 28, 2024 9:26 pm I think part of it is that more recently Google has added AI which, when you ask, tries to summarize to give you an answer. Unfortunately you can't always trust the answer (*** example below), but for some, if they like that feature and accept it, it's easier than sifting through search results (I find my students doing this more and more now) I would think maybe eventually as search changes with AI expected as an integral part of it, this could also proliferate to forums like this.

My point was more like why can't the search engine providers give the option of simple EXACT MATCHES. Fuzzy searches and AI results have their uses but they rarely give good results.

pjw wrote: Wed Dec 25, 2024 11:36 pm Why not just use WMON ,x, y instead of all those lines of code?

Hey PJW...

Nice tip

Tried it and it works good!!

Although I hardly ever use a bigger resolution.

sitting lazy almost lying on a comfortable chair ..... an a distance from the QL
The "old" QLscreen is very readable

NL_QL_Usr wrote: Sun Dec 29, 2024 10:22 am

pjw wrote: Wed Dec 25, 2024 11:36 pm Why not just use WMON ,x, y instead of all those lines of code?

Hey PJW...

Nice tip

Tried it and it works good!!

Although I hardly ever use a bigger resolution.

sitting lazy almost lying on a comfortable chair ..... an a distance from the QL
The "old" QLscreen is very readable

When I first got a nice 26" HDMI monitor for my main computer some years
ago I used to go full 1920x1080. What a luxury to have that much space to
frolic in! My laptop was a Macbook Pro 13" with a 1280x800 screen. That too
seemed perfect. But then, after 10 years, I had to retire the Macbook and
buy a new laptop. I got a 13" HDMI Huwawei. 1920x1080 on a 13" was just too
small to read. Even on the large monitor my eyesight started to struggle. I
didnt like any of the interpolated resolutions, with or without fancy
dithering filters. So I ended up with 960x540 (1920/2=960, 1080/2 = 540),
full screen, on both monitors. No interpolation, no filtering.

QPC2 and SMSQmulator @ 960x540 on HDMI is crisp and clear and, while not as
spacious as full HDMI, almost fits two full page editor windows side by
side and a lot more besides. From my laptop I can work on both machines
using Windows Remote Desktop, switching between them pretty seamlessly and
cutting and pasting (and thanks to Martin's excellent networking software!
copying) between QPC2 on the main computer and SMSQmulator on the laptop.
My MINIMIZE button ensures quick and easy access to the host OS when I need
it (I use the identical setup on RasPI).

One man's setup is, Im sure, as good as another's, but unless Im made aware
of a better one, I prefer my own.

PS: In case it wasnt clear. The resolutions Im talking about relate only to the
emulators. Windows and Linux are, of course, running at full HDMI.

Unfortunately you can't always trust the answer

Because it isn't AI, its just regurgitating what some idiot wrote!

XorA wrote: Sun Dec 29, 2024 6:26 pm

Unfortunately you can't always trust the answer

Because it isn't AI, its just regurgitating what some idiot wrote!

It is AI in the traditional sense where you basically create an expert system that looks up facts and tries to deduce them. The problem is, people have confused what AI is with real intelligence since these massive AI language models can sound more,and more real, even though sometimes the facts they spew are incorrect. It reminds me of a recent post on the QL Facebook group where someone touts a mid 80s Sinclair QL article on AI and proudly states how advanced the QL was back then. Well, we had chess games prior to that, as just one example. In fact, that same year Tim Hartnell came out with his book "Artificial Intelligence on the QL" and in one example he goes into how you can use matches in a matchbook to simulate learning, since it's just a weight-based system.

I think the one interesting element is, if you query this expert system enough, the inferencing of "what idiots say" gets better. There was a study that found that if you have people independently guessing something (I think this particular study was guessing the weight of a cow), that the average of all guesses was pretty close to being accurate) and so that's kind of the power of these massive language model systems.

bwinkel67 wrote: Sun Dec 29, 2024 9:21 pm

XorA wrote: Sun Dec 29, 2024 6:26 pm

Unfortunately you can't always trust the answer

Because it isn't AI, its just regurgitating what some idiot wrote!

It is AI in the traditional sense where you basically create an expert system that looks up facts and tries to deduce them. The problem is, people have confused what AI is with real intelligence since these massive AI language models can sound more,and more real, even though sometimes the facts they spew are incorrect. It reminds me of a recent post on the QL Facebook group where someone touts a mid 80s Sinclair QL article on AI and proudly states how advanced the QL was back then. Well, we had chess games prior to that, as just one example. In fact, that same year Tim Hartnell came out with his book "Artificial Intelligence on the QL" and in one example he goes into how you can use matches in a matchbook to simulate learning, since it's just a weight-based system.

I think the one interesting element is, if you query this expert system enough, the inferencing of "what idiots say" gets better. There was a study that found that if you have people independently guessing something (I think this particular study was guessing the weight of a cow), that the average of all guesses was pretty close to being accurate) and so that's kind of the power of these massive language model systems.

The trouble is the training sets for these LLMs are well known to be utterly broken. Like they train them on AO3 and wonder why they all become nazi sex offenders.........

Its the ultimate shit in -> shit out system, and I don't think the people running them care enough to fix that! They just want their daily mail style clickbaits!

Would you guys please stop messing up this thread? There is a perfectly good thread elsewhere to discus AI etc.
Perhaps a moderator can shovel these posts to where they belong?

I have had some thoughts on pjw's posts, such as :-

One man's setup is, i'm sure, as good as another's, but unless i'm made aware
of a better one, I prefer my own

My setups come from adapting a high resolution SMS2 system I used a long time ago ago.
(see viewtopic.php?t=3646&start=80)

Qjump limited based in Cambridge was commissioned to develop SMS2 for Atari STs more than 30 years ago. It was designed to be easy to use and run at at resolution of 1024 x 768 in four colours on an Atari Mega ST and 640 x480 in B&W on an Atari ST.
Nearly everything that makes up the user interface that I use in various SMSQE systems comes from that period. Things like Hotkeys (including the sleep facility), Buttons, Qpac2, Things, etc.etc were developed by Qjump.

Users of SMS2 were not given a choice of interface. What is now known as the extended environment, together with Hotkeys etc. was built into SMS2. SMS2 was supplied in an active, instant-on, modifiable cartridge.
Users had no choice but to learn how to use this high resolution windowing system.
There was no possibility of using a QDOS type interface because the Qdos Basic interpreter was replaced with an interactive Sbasic development environment consisting of various componnts that acted together as one. There was no #0,#1 or #2 to damage.
It is this system that I use in SMSQE to write SBASIC programs. With this system one has no choice but to write windowing, multi tasking programs that run in high resolution systems.

As far as I am concerned the higher the display resolution the better. In my opinion Display resolution is much more useful than high speed. When I was using an Atari Mega STe, I never thought it was too slow.

The following picture shows QPC running SMSQE and Steem running SMS2. As you can see they are almost identical. From my perspective the main change since the SMS2 days, is the ability to run SMSQE at high resolutions on PCs, PIs and FPGA. With the introduction of FPGA for QL, any QL enthusiast has access to high resolution.