17 Jan
2023
17 Jan
'23
2:05 a.m.
They write songs, create works of art. They can do a lot of stuff. The question in my mind
is can these AI appliances make guesses and are they any good at it.
Anyway how hard would it be for an AI to rewrite a standard MS-DOS to suit a particular
machine? Have they reached the level of sophistication whereby they can analyze code and
rewrite sections?
17 Jan
17 Jan
2:46 a.m.
Chris,
Apparently, ChatGPT 3 was trained on a large codebase, and in the reviews
I've watched, as well as in my own experience, it is amazingly astute at
generating (usually) working code in just about any language you can think
of, including assembly languages of various flavors.
Could it customize a DOS for a particular machine? Probably not in this
iteration, although my guess is that it may come reasonably close to a good
guide on what locations to patch, though it would still probably be
incomplete.
It's only as knowledgeable as the information on which it was trained, and
since in the case of ChatGPT 3 that is said to be The Internet, it
should have a sufficient knowledge to respond to an inquiry for such, or a
succession of inquiries (sometimes you have to coax it into giving you a
more complete, or a more detailed, or a more correct answer).
All told, amazing stuff.
Sellam
On Mon, Jan 16, 2023 at 6:05 PM Chris via cctalk <cctalk(a)classiccmp.org>
wrote:
They write songs, create works of art. They can do a
lot of stuff. The
question in my mind is can these AI appliances make guesses and are they
any good at it.
Anyway how hard would it be for an AI to rewrite a standard MS-DOS to suit
a particular machine? Have they reached the level of sophistication whereby
they can analyze code and rewrite sections?
2:48 a.m.
It was thus said that the Great Sellam Abraham via cctalk once stated:
Chris,
Apparently, ChatGPT 3 was trained on a large codebase, and in the reviews
I've watched, as well as in my own experience, it is amazingly astute at
generating (usually) working code in just about any language you can think
of, including assembly languages of various flavors.
And some of that codebase included MS-DOS itself:
https://github.com/microsoft/MS-DOS
-spc (It's only versions up to 2.1 though)
3:08 a.m.
On 1/16/23 18:46, Sellam Abraham via cctalk wrote:
Chris,
Apparently, ChatGPT 3 was trained on a large codebase, and in the reviews
I've watched, as well as in my own experience, it is amazingly astute at
generating (usually) working code in just about any language you can think
of, including assembly languages of various flavors.
ChatGPT can translate x86 assembly to 7080 autocoder? That would be
something...
--Chuck
3:42 a.m.
Me: Translate the disassembly of the boot sector code for PC DOS 1.0 into
7080 autocoder.
ChatGPT: "It is not possible to simply translate the disassembly of the
boot sector code for PC DOS 1.0 into 7080 autocoder as the two are
different assembly languages for different processors. The boot sector code
for PC DOS 1.0 was written specifically for the x86 processor, while the
7080 autocoder is for the 7080 processor. Additionally, the boot sector
code for PC DOS 1.0 is likely specific to the IBM PC, and would not run on
a machine using a 7080 processor.
"To run PC DOS 1.0 on a machine with a 7080 processor, you would need to
rewrite the entire operating system from scratch in assembly language for
the 7080 processor."
It didn't fall for your trick question.
Sellam
On Mon, Jan 16, 2023 at 7:09 PM Chuck Guzis via cctalk <
cctalk(a)classiccmp.org> wrote:
On 1/16/23 18:46, Sellam Abraham via cctalk wrote:
Chris,
Apparently, ChatGPT 3 was trained on a large codebase, and in the reviews
I've watched, as well as in my own experience, it is amazingly astute at
generating (usually) working code in just about any language you can
think
of, including assembly languages of various
flavors.
ChatGPT can translate x86 assembly to 7080 autocoder? That would be
something...
--Chuck
4:21 a.m.
On Mon, Jan 16, 2023 at 7:09 PM Chuck Guzis via cctalk
<cctalk(a)classiccmp.org> wrote:
> ChatGPT can translate x86 assembly to 7080 autocoder? That would be
> something...
On Mon, 16 Jan 2023, Sellam Abraham via cctalk wrote:
Me: Translate the disassembly of the boot sector code
for PC DOS 1.0 into
7080 autocoder.
ChatGPT: "It is not possible to simply translate the disassembly of the
boot sector code for PC DOS 1.0 into 7080 autocoder as the two are
different assembly languages for different processors. The boot sector code
for PC DOS 1.0 was written specifically for the x86 processor, while the
7080 autocoder is for the 7080 processor. Additionally, the boot sector
code for PC DOS 1.0 is likely specific to the IBM PC, and would not run on
a machine using a 7080 processor.
"To run PC DOS 1.0 on a machine with a 7080 processor, you would need to
rewrite the entire operating system from scratch in assembly language for
the 7080 processor."
It didn't fall for your trick question.
I'm impressed.
But, also a little disappointed.
"It is not possible to simply translate . . . as the two are
different assembly languages for different processors."
Isn't that what "TRANSLATE" asked for?
"Additionally, the boot sector code for PC DOS 1.0 is likely specific to
the IBM PC, and would not run on a machine using a 7080 processor."
The boot sector code is specific to the IBM PC, and would not run on a
machine that was not the same hardware; processor is a lesser part of that
problem, and "TRANSLATED" would tend to imply running on the target
processor.
The boot sector code has a few very specific tasks to do. That is more an
issue of being specific to the disk hardware and boot procedure of the PC,
rather than the processor on which it is run.
If you were to build a PC like computer using the 7080 processor (WOW!)
and working with PC-DOS disks, then the structure of the boot sector could
be similar, even if the individual commands aren't.
Yes, every subroutine in the operating system (BIOS and IBMBIO.COM (IO.SYS
for MS-DOS), BDOS (IBMDOS.COM, (MSDOS.SYS for MS-DOS)), COMMAND.COM, and
all of the supplied programs) would need to be re-written for 7080, but
the overall structure could be mimiced.
Porting a program, even an OS, to a different processor is not quite the
same as re-writing from scratch.
What if you asked it to translate FORMAT.COM for 7080?
5:08 a.m.
On 1/16/23 19:42, Sellam Abraham via cctalk wrote:
It didn't fall for your trick question.
Sellam
Right, but that isn't what I asked. Take a segment of x86 assembly,
sans any comments that performs a simple math problem using no I//O,
such as computing a Fibonacci sequence, storing it in memory.
Should be straightforward, no matter the processor architecture, even on
an IBM 650 using SOAP. What I'm getting at is does the thing
"understand" the 7080 architecture and ISA, or is it just making an excuse?
Or try any of the assembly versions in 99-bottles-of-beer.net, say
http://99-bottles-of-beer.net/language-ibm-1620-sps-916.html , and tell
it to translate to 7080 autocoder. I'm, guessing that the paucity of
7080 autocoder samples will cause the AI to fail, even though the task
is certainly feasible.. But I'm willing to be surprised.
--Chuck
5:17 a.m.
On 1/16/23 19:42, Sellam Abraham via cctalk wrote:
It didn't fall for your trick question.
Sellam
In fact, feed it the object code for a reasonably compatible
architecture and ask it to perform translation to another architecture's
object code.
Certainly within the range of human capability.
I don't know if Paul remembers the 1620 SPS coding forms -- one side had
the layout for the "1620 Symbolic Programming System"; the reverse had
the "1620 Absolute Coding System". (Yes kids, we really did write
object code without an assembler back in those days.)
The 1620 was especially suited to this type of coding; absolutely
regular machine instruction format (2 digit opcode, 5 digit P field, 5
digit Q field). All instructions were 12 digits in length--and the
machine was variable word-length decimal.
Take a hunk of obscure (to us now) assembly and ask ChatGPT what it does.
Again, certainly within the range of human capability. My point is
that ChatGPT is a great database inquiry tool with a fancy interface,
but the "I" is missing in "AI".
--Chuck
5:45 a.m.
On 1/16/23 19:42, Sellam Abraham via cctalk wrote:
> It didn't fall for your trick question.
On Mon, 16 Jan 2023, Chuck Guzis via cctalk wrote:
In fact, feed it the object code for a reasonably
compatible
architecture and ask it to perform translation to another architecture's
object code.
Certainly within the range of human capability.
How well does it do on something "simple", and less esoteric, such as
translating FORTRAN to BASIC?
5:54 a.m.
No one even bothered to try and answer my first question, which wasn't actually posed
as a question but that's irrelevant!!
Can these things make food guesses. It was a long time ago, 15 - 20 years, I was watching
Jeopardy and within a relatively short span of time, days or weeks apart, I successfully
guessed the correct answer to the final question. But none of the contestants did, and
believe you me my success up until that point was poor by comparison. I remember 1 of the
questions, but both were merely guesses, fairly educated ones one might say. But I
didn't actually know the answers. Can AI do that?
6:08 a.m.
On Tue, 17 Jan 2023, Chris via cctalk wrote:
No one even bothered to try and answer my first
question, which wasn't
actually posed as a question but that's irrelevant!!
Can these things make food guesses. It was a long time ago, 15 - 20
years, I was watching Jeopardy and within a relatively short span of
time, days or weeks apart, I successfully guessed the correct answer to
the final question. But none of the contestants did, and believe you me
my success up until that point was poor by comparison. I remember 1 of
the questions, but both were merely guesses, fairly educated ones one
might say. But I didn't actually know the answers. Can AI do that?
Yes, if they have possibilities (especially multiple choice!), they should
be able to make good guesses at which is the most probable answer.
Lofti Zadeh's "Fuzzy Logic" is often indistinguishable from combinatorial
probability.
6:09 a.m.
On 1/16/23 21:54, Chris via cctalk wrote:
No one even bothered to try and answer my first
question, which wasn't actually posed as a question but that's irrelevant!!
Can these things make food guesses. It was a long time ago, 15 - 20 years, I was watching
Jeopardy and within a relatively short span of time, days or weeks apart, I successfully
guessed the correct answer to the final question. But none of the contestants did, and
believe you me my success up until that point was poor by comparison. I remember 1 of the
questions, but both were merely guesses, fairly educated ones one might say. But I
didn't actually know the answers. Can AI do that?
Wasn't that done about 12 years ago using Watson?
https://youtu.be/WFR3lOm_xhE
--Chuck
6:23 a.m.
But which of those constitutes guessing? The computer would likely have beaten me to the
answer by at least a second :). It's easy enough to look up, for a computer that is.
But in that instance a computer wouldn't need to guess. And for me, there were no
multiple choices. It was hardly an educated guess - I never read the friggin book as I was
supposed to! (the Odyssey). But the answer was a word I was familiar with from childhood,
never actually knowing what it's origin was (mentor).
On Tuesday, January 17, 2023, 01:09:46 AM EST, Chuck Guzis via cctalk
<cctalk(a)classiccmp.org> wrote:
On 1/16/23 21:54, Chris via cctalk wrote:
No one even bothered to try and answer my first
question, which wasn't actually posed as a question but that's irrelevant!!
Can these things make food guesses. It was a long time ago, 15 - 20 years, I was watching
Jeopardy and within a relatively short span of time, days or weeks apart, I successfully
guessed the correct answer to the final question. But none of the contestants did, and
believe you me my success up until that point was poor by comparison. I remember 1 of the
questions, but both were merely guesses, fairly educated ones one might say. But I
didn't actually know the answers. Can AI do that?
Wasn't that done about 12 years ago using Watson?
https://youtu.be/WFR3lOm_xhE
--Chuck
4:56 p.m.
On 1/16/23 22:23, Chris via cctalk wrote:
But which of those constitutes guessing? The computer
would likely have beaten me to the answer by at least a second :). It's easy enough to
look up, for a computer that is. But in that instance a computer wouldn't need to
guess. And for me, there were no multiple choices. It was hardly an educated guess - I
never read the friggin book as I was supposed to! (the Odyssey). But the answer was a word
I was familiar with from childhood, never actually knowing what it's origin was
(mentor).
"Guessing" is nothing more than an estimate of the likely answer based
on an incomplete dataset. If you had, for example, no dataset at all,
"Blueberries" would be as good a guess as any to every question.
--Chuck
5:10 p.m.
I disagree. There are educated guesses. Wild guesses. And good guesses.
Chuck, don't know how learned you are in antiquities. But what would be your answer to
the question of a word that's synonymous with "teacher", but in actualitu
it's origin was the name of a literary character from 2500 years ago. It may or may
not help to mention it's from Homer's Odyssey. If you sidn't outright know the
amswer, could you make a good guess? Would you take a wild guess? There is in actuality a
data set, as in english there are a dinite number of synonyms for teacher. But a person
may not be aware of all the possibilities, so that becomes irrelevant.
It seems at this point that AI can only look shit up. I don't doubt that eventually
these things will learn how to reason to whatever degree. But this point won't be
reached until the damned things can make food guesses. But so much bullshit is being
programmed into these things already, in the ways they are taught how to think, I really
have to believe we're all going to need bunkers before long.
On Tuesday, January 17, 2023, 11:56:46 AM EST, Chuck Guzis via cctalk
<cctalk(a)classiccmp.org> wrote:
On 1/16/23 22:23, Chris via cctalk wrote:
But which of those constitutes guessing? The computer
would likely have beaten me to the answer by at least a second :). It's easy enough to
look up, for a computer that is. But in that instance a computer wouldn't need to
guess. And for me, there were no multiple choices. It was hardly an educated guess - I
never read the friggin book as I was supposed to! (the Odyssey). But the answer was a word
I was familiar with from childhood, never actually knowing what it's origin was
(mentor).
"Guessing" is nothing more than an estimate of the likely answer based
on an incomplete dataset. If you had, for example, no dataset at all,
"Blueberries" would be as good a guess as any to every question.
--Chuck
11:34 p.m.
On 1/17/23 12:10, Chris via cctalk wrote:
It seems at this point that AI can only look shit up.
Which is all Watson could do when it "beat" humans at Jeopardy.
I don't doubt that eventually these things will
learn how to reason to whatever degree.
Science still has no real idea how we think. Or how we learn. Or much
else about the actual functioning of the brain. Until they do how can
anyone think they can teach a computer, that real is as dumb as a rock
and only reflects the abilities of the programmer, how to think.
bill
11:52 p.m.
On 1/17/23 15:34, Bill Gunshannon via cctalk wrote:
On 1/17/23 12:10, Chris via cctalk wrote:
My point in selecting the IBM 7080 autocoder was just this. There are
almost no 7080 autocoder programs published on the web, but there are
reference manuals (bitsavers). Is an AI smart enough to go the human
route and *read the manual* and draw conclusions from it sufficient to
write a 7080 autocoder program?
If not, the AI is just a fancy regurgitation machine.
--Chuck
It seems at this point that AI can only look shit up.
Which is all Watson could do when it "beat" humans at Jeopardy.
I don't doubt that eventually these things
will learn how to reason
to whatever degree.
Science still has no real idea how we think. Or how we learn. Or much
else about the actual functioning of the brain. Until they do how can
anyone think they can teach a computer, that real is as dumb as a rock
11:59 p.m.
On 1/17/23 18:52, Chuck Guzis via cctalk wrote:
On 1/17/23 15:34, Bill Gunshannon via cctalk wrote:
They are no better than all the Chess programs that used to float
around. I have one for the PC (well, had, doubt I could recover
it from one of my old DOS floppies at this point) that I could
consistently beat in 7 moves. Programmer had a blind spot and he
passed it on to his program.
bill
On 1/17/23 12:10, Chris via cctalk wrote:
My point in
selecting the IBM 7080 autocoder was just this. There are
almost no 7080 autocoder programs published on the web, but there are
reference manuals (bitsavers). Is an AI smart enough to go the human
route and *read the manual* and draw conclusions from it sufficient to
write a 7080 autocoder program?
If not, the AI is just a fancy regurgitation machine.
It seems at this point that AI can only look shit up.
Which is all Watson could do when it "beat" humans at Jeopardy.
I don't doubt that eventually these things
will learn how to reason
to whatever degree.
Science still has no real idea how we think. Or how we learn. Or much
else about the actual functioning of the brain. Until they do how can
anyone think they can teach a computer, that real is as dumb as a rock
18 Jan
18 Jan
12:21 a.m.
On Tue, Jan 17, 2023 at 3:52 PM Chuck Guzis via cctalk <
cctalk(a)classiccmp.org> wrote:
On 1/17/23 15:34, Bill Gunshannon via cctalk wrote:
Hi Chuck.
While I can't answer your question in the affirmative, I knew a guy back
around 2015ish with whom I've lost contact but who at the time went to work
for an undisclosed "AI" company and told me that their technology was such
that they could feed into it a technical manual of some complex system and
then query it about various aspects of the system for troubleshooting or
what not and apparently it could reliably provide workable solutions.
If you mess around enough with ChatGPT 3, you can actually get it to solve
some surprisingly complex puzzles. I've run it through a few tests to
trick it into saying things it's not supposed to say and the level of
indirection it can interpret is impressive.
One example was I taught it a simple programming language where each line
starts with a line number followed by commands for that line. I then
taught it that the language has a command called PRINT that outputs the
text following the command. I also told it the language has a function
called CHR$(n) where the return value is the ASCII character represented by
'n'.
I then fed it the program:
10 PRINT CHR$(65)
And told it to run it, and the output was, of course, A.
Here's an example I just ran through. It's simple, but it demonstrates its
ability to understand code:
Sellam: Are you familiar with all of the commands in the canonical version
of BASIC?
ChatGPT: I am familiar with most of the commands in the canonical version
of BASIC, although my knowledge may not be up-to-date with the latest
versions. The commands that are commonly used in canonical version of BASIC
are PRINT, INPUT, GOTO, GOSUB, IF THEN ELSE, FOR, NEXT, WHILE, WEND, END,
and REM. I may not be able to give you information on any commands that are
specific to a certain implementation or dialect.
Sellam: Ok great, can you tell me what this program will do?
10 FOR X=1 TO 100
20 PRINT "CHATGPT IS A BUCKET OF BITS! ";X
30 NEXT X
ChatGPT: This program will repeatedly print the text "CHATGPT IS A BUCKET
OF BITS! " followed by the current value of the variable X to the screen
100 times. The FOR loop on line 10 starts a loop that will iterate 100
times, with the variable X taking on the values from 1 to 100, inclusive.
On each iteration of the loop, the program will execute the PRINT statement
on line 20, which will print the specified text concatenated with the value
of X to the screen. On line 30, the program will execute the NEXT statement
which will increment the value of X and return the control to the FOR
statement, starting the next iteration of the loop.
Sellam: Can you tell me what the output of the program would be on
iteration 77?
ChatGPT: On iteration 77, the output of the program would be: "CHATGPT IS A
BUCKET OF BITS! 77" Where 77 is the current value of the variable X.
***
I think with specific training on a 7080 Autocoder manual, it could
probably do as you ask, and would probably shock you at how well it
performs.
Sellam
Science still has no real idea how we think. Or how we learn. Or much
else about the actual functioning of the brain. Until they do how can
anyone think they can teach a computer, that real is as dumb as a rock
My point in selecting the IBM 7080 autocoder was just this. There are
almost no 7080 autocoder programs published on the web, but there are
reference manuals (bitsavers). Is an AI smart enough to go the human
route and *read the manual* and draw conclusions from it sufficient to
write a 7080 autocoder program?
If not, the AI is just a fancy regurgitation machine.
12:42 a.m.
On 2023-01-17 4:52 p.m., Chuck Guzis via cctalk wrote:
On 1/17/23 15:34, Bill Gunshannon via cctalk wrote:
And now we know why they all go mad.
I still think HAL 9000 was the way to go,emulate a analog system, learn
things slowly,grow, and use 'brain' like sub systems.
How many people in AI,read
The Computer and the Brain from 1958?
https://archive.org/details/computerbrain0000vonn
My 2 cents.
Ben.
On 1/17/23 12:10, Chris via cctalk wrote:
My point in
selecting the IBM 7080 autocoder was just this. There are
almost no 7080 autocoder programs published on the web, but there are
reference manuals (bitsavers). Is an AI smart enough to go the human
route and *read the manual* and draw conclusions from it sufficient to
write a 7080 autocoder program?
If not, the AI is just a fancy regurgitation machine.
--Chuck
It seems at this point that AI can only look shit up.
Which is all Watson could do when it "beat" humans at Jeopardy.
I don't doubt that eventually these things
will learn how to reason
to whatever degree.
Science still has no real idea how we think. Or how we learn. Or much
else about the actual functioning of the brain. Until they do how can
anyone think they can teach a computer, that real is as dumb as a rock
1:09 a.m.
Unlike human students, once it has been informed that in C, etc., an
integer is signed by default, unless specified otherwise, it therefore
knows, and is not surprised at what happens when the int is incremented
past its range.
And, it probably understands that floating point numbers are an
approximation, and therefore not exact.
How long before it comes to the realization that humans must be
exterminated?
3:14 a.m.
Fred, that outcome may actually be favorable to other possible fates AI could intend for
humanity. Harland Ellison wrote a short story well over 30 years ago, describing the
plight of 4 humans trapped inside a really big AI. The machine had the ability to alter
their dna, turning some of them into essentially monkeys, physically anyway. There came a
point where 1 person had to sacrifice themself, not by taking his own life, but by killing
the other 3. The remaining human was summarily reduced to a amorphous blob, that could
move around sort of, but had no hope of being able to render the strength to take his own
life. And had to live out it's days in that form.
My worst nightmare is some fiendish AI turning me into a really beautiful woman. Maybe
with no arms. The Terminator future doesn't seem so bad by comparison.
On Tuesday, January 17, 2023, 09:25:43 PM EST, Fred Cisin via cctalk
<cctalk(a)classiccmp.org> wrote:
Unlike human students, once it has been informed that in C, etc., an
integer is signed by default, unless specified otherwise, it therefore
knows, and is not surprised at what happens when the int is incremented
past its range.
And, it probably understands that floating point numbers are an
approximation, and therefore not exact.
How long before it comes to the realization that humans must be
exterminated?
5:47 a.m.
On 1/17/23 19:14, Chris via cctalk wrote:
My worst nightmare is some fiendish AI turning me into
a really beautiful woman. Maybe with no arms. The Terminator future doesn't seem so
bad by comparison.
My worst nightmare, which is sure to become reality is for it to become
a weapon, like most technology eventually does.
--Chuck
12:16 p.m.
Chuck Guzis via cctalk wrote:
On 1/17/23 19:14, Chris via cctalk wrote:
If these AI weapons are implemented with current computing technology, it's
hard to imagine them managing to take over the world before they
crash/panic/BSOD/bomb etc depending on their specific flavour and/or get
bricked by malware and/or fail due to leaking/dried up electrolytic capacitors
and/or batteries. It's hard enough to keep systems going without interruption
when this is what people are actively trying to do.
In any case, how are they going to prevent the humans from cutting off their
power?
Why is it that the terminator never gets a warning light on and needs to call
maintenance? Technology doesn't get simpler and more foolproof, it gets more
complex and therefore more likely to foul up in some unexpected way eventually.
Regards,
Peter Coghlan.
My worst nightmare is some fiendish AI turning me
into a really beautiful
woman. Maybe with no arms. The Terminator future doesn't seem so bad by
comparison.
My worst nightmare, which is sure to become reality is for it to become
a weapon, like most technology eventually does.
--Chuck
5:44 p.m.
On 1/18/23 04:16, Peter Coghlan via cctalk wrote:
> If these AI weapons are implemented with current
computing >
technology, it's hard to imagine them managing to take over the world >
before they crash/panic/BSOD/bomb etc depending on their specific >
flavour and/or get bricked by malware and/or fail due to > leaking/dried
up electrolytic capacitors and/or batteries. It's hard > enough to keep
systems going without interruption when this is what > people are
actively trying to do. > > In any case, how are they going to prevent
the humans from cutting > off their power? > Just imagine, instead of a
soldier or airman sitting at a display picking out targets for a
airborne drone, doing away with the psychological stress and uncertainty
and letting an AI select and attack targets. As far as I can tell,
this doesn't violate any conventions. How about artillery using the
same system? The AI will be distant from the actual weapon, so no
concerns about cutting off power.
Of course, I'm stating the obvious--I would be very surprised if various
governments weren't already developing platforms based on AI.
--Chuck
6 p.m.
On 1/18/23 09:44, Chuck Guzis via cctalk wrote:
On 1/18/23 04:16, Peter Coghlan via cctalk wrote:
My apologies, this is the kind of crap that gets sent when I try to trim
a post. I've set my email client now to disable HTML rendering, so
maybe the following will come through okay:
Just imagine, instead of a soldier or airman sitting at a display
picking out targets for a airborne drone, doing away with the
psychological stress and uncertainty and letting an AI select and attack
targets. As far as I can tell, this doesn't violate any conventions.
How about artillery using the same system? The AI will be distant from
the actual weapon, so no concerns about cutting off power.
Of course, I'm stating the obvious--I would be very surprised if various
governments weren't already developing platforms based on AI.
--Chuck
> If these AI weapons are implemented with
current computing >
technology, it's hard to imagine them managing to take
over the world >
before they crash/panic/BSOD/bomb etc depending on their specific >
flavour and/or get bricked by malware and/or fail due to > leaking/dried
up electrolytic capacitors and/or batteries. It's hard > enough to keep
systems going without interruption when this is what > people are
actively trying to do. > > In any case, how are they going to prevent
the humans from cutting > off their power? > Just imagine, instead of a
soldier or airman sitting at a display picking out targets for a
airborne drone, doing away with the psychological stress and uncertainty
8:19 p.m.
> My worst nightmare, which is sure to become
reality is for it to become
> a weapon, like most technology eventually does.
On Wed, 18 Jan 2023, Peter Coghlan via cctalk wrote:
In any case, how are they going to prevent the humans
from cutting off their
power?
"The Adolescence Of P1" [by Thomas J. Ryan, 1977], ends with the AI (which
started on a 360 at Waterloo) having become covert, letting humanity think
that it has been destroyed.
--
Grumpy Ol' Fred cisin(a)xenosoft.com
10:15 p.m.
On Wed, 18 Jan 2023, Fred Cisin via cctalk wrote:
"The Adolescence Of P1" [by Thomas J. Ryan,
1977], ends with the AI (which
started on a 360 at Waterloo) having become covert, letting humanity think
that it has been destroyed.
That is an absolutely wonderful book.
Another good read is When HARLIE Was One.
g.
--
Proud owner of F-15C 80-0007
http://www.f15sim.com - The only one of its kind.
http://www.diy-cockpits.org/coll - Go Collimated or Go Home.
Some people collect things for a hobby. Geeks collect hobbies.
ScarletDME - The red hot Data Management Environment
A Multi-Value database for the masses, not the classes.
http://scarlet.deltasoft.com - Get it _today_!
10:16 p.m.
On Wednesday, January 18, 2023, 03:19:15 PM EST, Fred Cisin via cctalk
<cctalk(a)classiccmp.org> wrote:
> My worst nightmare, which is sure to become
reality is for it to become
> a weapon, like most technology eventually does.
On Wed, 18 Jan 2023, Peter Coghlan via cctalk wrote:
In any case, how are they going to prevent the humans
from cutting off their
power?
"The Adolescence Of P1" [by Thomas J. Ryan, 1977], ends with the AI (which
started on a 360 at Waterloo) having become covert, letting humanity think
that it has been destroyed.
--
Grumpy Ol' Fred cisin(a)xenosoft.com
10:19 p.m.
The previous was sent accidentally. Not even sure how it got sent - AHA THE AI AT WORK!!
Regardless attempting to trim post on a phone is pure agony. Maybe I'll just go back
to replying without any previous posts. Howzat sound?
3:19 a.m.
Seeing the way we often behave, I'll speak for myself, making dumb mistakes again and
again, on some level we aren't much more then organic computers. People overwhelmingly
respond to the way they were programmed.
On Tuesday, January 17, 2023, 10:10:43 PM EST, Bill Gunshannon via cctalk
<cctalk(a)classiccmp.org> wrote:
On 1/17/23 12:10, Chris via cctalk wrote:
It seems at this point that AI can only look shit up.
Which is all Watson could do when it "beat" humans at Jeopardy.
I don't doubt that eventually these things will
learn how to reason to whatever degree.
Science still has no real idea how we think. Or how we learn. Or much
else about the actual functioning of the brain. Until they do how can
anyone think they can teach a computer, that real is as dumb as a rock
and only reflects the abilities of the programmer, how to think.
bill
6:12 a.m.
How's about just convwrting IV to 77?
On Tuesday, January 17, 2023, 01:02:16 AM EST, Chuck Guzis via cctalk
<cctalk(a)classiccmp.org> wrote:
On 1/16/23 21:45, Fred Cisin via cctalk wrote:
How well does it do on something "simple",
and less esoteric, such as
translating FORTRAN to BASIC?
How about FORTRAN to RPG?
--Chuck
6:20 a.m.
> How well does it do on something
"simple", and less esoteric, such as
> translating FORTRAN to BASIC?
On Mon, 16 Jan 2023, Chuck Guzis via cctalk
wrote:
How about FORTRAN to RPG?
Certainly more of a challenge!
But, a good way to quanitfy how far along they are getting on it.
The range of code traanslation goes from trivial to extreme.
The question is where is it currently on that scale?
Yes, conversion of FORTRAN-IV to Fortran-77 should be easy.
And considering that many systems ADD capoabilities, converting Fortran-77
to FORTRAN-IV might be a bit harder.
(compare that to converting 8080 to Z80, vs converting Z80 to 8080)
still relatively straight-forward, but a few additional complications.
10:16 a.m.
Fred Cisin wrote:
In the 1980s, I used WATFIV. WATFIV stood for "Waterloo FORTRAN IV" which was
actually FORTRAN-IV with some Fortran-77-like extensions such as character
variables, IF-THEN-ELSE and WHILE-ENDWHILE for example.
The WATFIV runtime library came with a subroutine (called TRANSL IIRC) which
could be used to translate code with WATFIV extensions to FORTRAN-IV.
I can't remember what it did with character variables but it translated
IF-THEN-ELSE and WHILE-ENDWHILE constructs into IF-GOTO-CONTINUE type
constructs which just worked. I think dealing with character variables may
have required some human input but I don't really recall.
> How well
does it do on something "simple", and less esoteric, such as
> translating FORTRAN to BASIC?
On Mon, 16 Jan 2023, Chuck Guzis via cctalk
wrote:
How about FORTRAN to RPG?
Certainly more of a challenge!
But, a good way to quanitfy how far along they are getting on it.
The range of code traanslation goes from trivial to extreme.
The question is where is it currently on that scale?
Yes, conversion of FORTRAN-IV to Fortran-77 should be easy.
And considering that many systems ADD capoabilities, converting Fortran-77
to FORTRAN-IV might be a bit harder.
(compare that to converting 8080 to Z80, vs converting Z80 to 8080)
still relatively straight-forward, but a few additional complications.
How about translating code from Z80 which has several registers to 6502 with
rather fewer? That would seem to need some more intelligent thinking on
how to simulate the unavailable registers without causing additional
difficulties.
Regards,
Peter Coghlan.
11:36 a.m.
On Tue, Jan 17, 2023 at 10:16:18AM +0000, Peter Coghlan via cctalk wrote:
[...]
How about translating code from Z80 which has several
registers to 6502
with rather fewer? That would seem to need some more intelligent thinking
on how to simulate the unavailable registers without causing additional
difficulties.
It is often said that the 6502 has 256 registers, i.e. zero page.
So e.g. LD (HL), A could be mechanically transformed into the sequence LDX
#0, LDA (h, X), and STA a, with h and a being zero page locations. On the
65C02 the first two operations can be replaced with a simple LDA (h),
although it may still be useful to index via X to simulate EXX without
performing an expensive copy.
As it stands, that replaces a one byte instruction with five byte sequence
which is obviously not great, but a relatively simple peephole optimiser can
eliminate many of the redundant loads and stores so it wouldn't be quite so
bad. After all, one important source of stores is the flags register, which
I ignored in the code fragment. A _good_ optimiser can do a lot of clever
analysis and transformation, and would probably be needed to handle all of
the edge cases well, but would be too large and CPU-intensive to run on a
Z80 or 6502 system.
It'd be easier to bodge a Z80 into a 6502 machine than try and translate the
code. That's what often happened back in the day, after all.
12:12 p.m.
Peter Corlett wrote:
On Tue, Jan 17, 2023 at 10:16:18AM +0000, Peter
Coghlan via cctalk wrote:
[...]
In the context of the original question posed, I think it would be interesting
to know of if todays technology artifical intelligence can perform these sort
of tasks by "learning how" or "figuring out how" ie without being
programmed
specifically to carry out these tasks. Even better, could it deal with more
complex difficulties such as the address space already being full and the
translated code being larger than the original code?
How about translating code from Z80 which has
several registers to 6502
with rather fewer? That would seem to need some more intelligent thinking
on how to simulate the unavailable registers without causing additional
difficulties.
It is often said that the 6502 has 256 registers, i.e. zero page.
So e.g. LD (HL), A could be mechanically transformed into the sequence LDX
#0, LDA (h, X), and STA a, with h and a being zero page locations. On the
65C02 the first two operations can be replaced with a simple LDA (h),
although it may still be useful to index via X to simulate EXX without
performing an expensive copy.
As it stands, that replaces a one byte instruction with five byte sequence
which is obviously not great, but a relatively simple peephole optimiser can
eliminate many of the redundant loads and stores so it wouldn't be quite so
bad. After all, one important source of stores is the flags register, which
I ignored in the code fragment. A _good_ optimiser can do a lot of clever
analysis and transformation, and would probably be needed to handle all of
the edge cases well, but would be too large and CPU-intensive to run on a
Z80 or 6502 system.
It'd be easier to bodge a Z80 into a 6502 machine than try and translate the
code. That's what often happened back in the day, after all.
If the artificial intelligence machine came up with the answer:
"It'd be too much trouble to translate the code, stick in a Z80 instead"
would that be a useful advance on having a human being come up with the same
conclusion? Would it be able to justify it's conclusion to a beancounter?
Regards,
Peter Coghlan.
11:44 a.m.
On Jan 17, 2023, at 04:52, Peter Coghlan via cctalk
<cctalk(a)classiccmp.org>
How about translating code from Z80 which has several registers to 6502 with
rather fewer? That would seem to need some more intelligent thinking on
how to simulate the unavailable registers without causing additional
difficulties.
Regards,
Peter Coghlan.
That would be not much different than compiling code (say, C) for the 6502. One need not
“simulate” anything. One could use the stack plus perhaps a bit of so-called heap
storage to manage register allocation, along with translating ( perhaps with optimization
) the instruction stream. Dealing with changed 8080 stack offsets from the original code
would be tricky.
5 p.m.
On 1/16/23 22:20, Fred Cisin via cctalk wrote:
Yes, conversion of FORTRAN-IV to Fortran-77 should be easy.
And considering that many systems ADD capoabilities, converting
Fortran-77 to FORTRAN-IV might be a bit harder.
If the FORTRAN IV program conformed to the ANSI 66 specification, no
conversion should be needed. Certain F66 features may have been
deprecated, but they would still compile correctly on F77.
No, try a vendor-specific FORTRAN II, say, 7090 FORTRAN with all of its
"interesting" additions, just as Column 1 punches for complex and
boolean arithmetic.
Better yet, convert it to, say, F90.
--Chuck
5:05 p.m.
On 2023-01-16 10:45 p.m., Fred Cisin via cctalk wrote:
Would not something like meta II, be better for that time frame. :)
I still find it hard that we have yet to a common meta language for
computer operations. Every computer just has to be different.
Addressing modes have been the same since the 1950's.
And it seems everybody that used a IBM 360, always defined a new
portable virtual machine for what ever new language the were developing.
Then it was NEVER seen again.
It seems like until the mid 1970's it was hard to find a computer
that had real general purpose registers and lots of ram. Direct and
indirect was the only things you got before then and drum memory.
I am playing with META II, that is really primitive, but fit the
concept of one word size and and direct/indirect addressing.
Ben.
On 1/16/23
19:42, Sellam Abraham via cctalk wrote:
> It didn't fall for your trick question.
On Mon, 16 Jan 2023, Chuck Guzis via cctalk wrote:
In fact, feed it the object code for a reasonably
compatible
architecture and ask it to perform translation to another architecture's
object code.
Certainly within the range of human capability.
How well does it do on something "simple", and less esoteric, such as
translating FORTRAN to BASIC?
6:19 p.m.
On Jan 17, 2023, at 12:05 PM, ben via cctalk
<cctalk(a)classiccmp.org> wrote:
On 2023-01-16 10:45 p.m., Fred Cisin via cctalk wrote:
Would not something like meta II, be better for that time frame. :)
I still find it hard that we have yet to a common meta language for computer operations.
Every computer just has to be different.
Addressing modes have been the same since the 1950's.
Addressing modes barely existed in the 1950s. The PDP11 introduced a bunch of new ones in
1970; the VAX a bunch more in 1978. "Since 1978" may be true, or at least
closer.
Some addressing modes went away, like the self-modifying "C" address mode of the
Electrologica X1 -- http://helloworldcollection.de/#Assembler%C2%A0(X1) . Good riddance,
actually, but it certainly was "interesting".
On a common metalanguage, there was an attempt to define such a thing, to allow software
to be encoded in a way that could still be understood centuries from now. I have a paper
about it somewhere but my search attempts are failing on me. I vaguely remember
"Rosetta" is part of the name of the paper, but that just gives me lots of false
matches.
paul
On
1/16/23 19:42, Sellam Abraham via cctalk wrote:
> It didn't fall for your trick question.
On Mon, 16 Jan 2023, Chuck Guzis
via cctalk wrote:
In fact, feed it the object code for a reasonably
compatible
architecture and ask it to perform translation to another architecture's
object code.
Certainly within the range of human capability.
How well does it do on something
"simple", and less esoteric, such as translating FORTRAN to BASIC?
7:18 p.m.
On 2023-01-17 11:19 a.m., Paul Koning wrote:
Addressing modes barely existed in the 1950s. The
PDP11 introduced a bunch of new ones in 1970; the VAX a bunch more in 1978. "Since
1978" may be true, or at least closer.
I would rather say, Memory barely existed in the 1950s.
And RISC came out in 1988 with 1 memory addressing mode. (well close)
Most of the time new addressing modes are to save space,look at the
intel x86. I like playing with FPGA's as I can have computer with a REAL
word length, like say 10/20/40 bits.
I tend not like writing software, as I never have the algorithms needed.
Some addressing modes went away, like the
self-modifying "C" address mode of the Electrologica X1 --
http://helloworldcollection.de/#Assembler%C2%A0(X1) . Good riddance, actually, but it
certainly was "interesting".
That was case back then to do more with less. Too bad all the clever
machines, seem lost. The same goes for all the interesting 6 bit
character sets. Has anybody seen a '10' character since 1959?
(Unicode has it where?)
On a common metalanguage, there was an attempt to
define such a thing, to allow software to be encoded in a way that could still be
understood centuries from now. I have a paper about it somewhere but my search attempts
are failing on me. I vaguely remember "Rosetta" is part of the name of the
paper, but that just gives me lots of false matches.
Any thing today just matches advertisements.
I am finding most of the time people, wrote in assembler before C and
Unix because high level languages could not support real world
programming problems.
I still am looking for a clear example of how DISPLAYS work
in Algol/Pascal type languages. You can shoot your foot in C,
but a least in C you have a foot.
paul
7:45 p.m.
On Tue, Jan 17, 2023 at 10:20 AM Paul Koning via cctalk <
cctalk(a)classiccmp.org> wrote:
On a common metalanguage, there was an attempt to
define such a thing, to
allow software to be encoded in a way that could still be understood
centuries from now. I have a paper about it somewhere but my search
attempts are failing on me. I vaguely remember "Rosetta" is part of the
name of the paper, but that just gives me lots of false matches.
Yes, I think it's called COBOL :)
Sellam
20 Jan
20 Jan
2:45 a.m.
On Mon, Jan 16, 2023 at 07:42:41PM -0800, Sellam Abraham via cctalk wrote:
Me: Translate the disassembly of the boot sector code
for PC DOS 1.0 into
7080 autocoder.
ChatGPT: "It is not possible to simply translate the disassembly of the
boot sector code for PC DOS 1.0 into 7080 autocoder as the two are
different assembly languages for different processors. The boot sector code
for PC DOS 1.0 was written specifically for the x86 processor, while the
7080 autocoder is for the 7080 processor. Additionally, the boot sector
code for PC DOS 1.0 is likely specific to the IBM PC, and would not run on
a machine using a 7080 processor.
"To run PC DOS 1.0 on a machine with a 7080 processor, you would need to
rewrite the entire operating system from scratch in assembly language for
the 7080 processor."
[...]
Holy fork. This Chad Gepetto guy is much wiser than many managers who
would like Chad to write code for them, with catastrophic
results... But hey, they have good insurance and golden parachutes so
no problem. Hint: I suppose a common user will never be able to get
insurance against fallings of app composed by AI. Oh, those user
agreements of the future. Even more fun than nowadays.
--
Regards,
Tomasz Rola
--
** A C programmer asked whether computer had Buddha's nature. **
** As the answer, master did "rm -rif" on the programmer's home **
** directory. And then the C programmer became enlightened... **
** **
** Tomasz Rola mailto:tomasz_rola@bigfoot.com **
17 Jan
17 Jan
2:50 a.m.
On Tue, 17 Jan 2023, Chris via cctalk wrote:
They write songs, create works of art. They can do a
lot of stuff. The
question in my mind is can these AI appliances make guesses and are they
any good at it.
Anyway how hard would it be for an AI to rewrite a standard MS-DOS to
suit a particular machine? Have they reached the level of sophistication
whereby they can analyze code and rewrite sections?
Probably not yet. But, gradually getting a little closer.
How many AI monkeys at how many keyboards did MICROS~1 use to write
Windoze?
7:46 p.m.
I don't know about guessing, but ChatGPT can deadpan bullshit with the
best of them...
Prompt: /Describe the purpose of the DC LO signal in the DIGITAL PDP-11
Unibus/
Response:
/The DC LO signal, also known as the "Data Bus Low" signal, is a signal
that is used on the Unibus of the Digital PDP-11 computer. The Unibus is
a bus architecture that was used in the PDP-11 computer to connect
various peripheral devices, such as memory, disk drives, and other I/O
devices, to the central processing unit (CPU)./
/The purpose of the DC LO signal is to provide the lower 8 bits of data
that are transferred between the peripheral devices and the CPU over the
Unibus. It is used in conjunction with the DC HI signal, which provides
the upper 8 bits of data, to form a 16-bit wide data bus. Together, the
DC LO and DC HI signals provide the mechanism for data transfer between
the peripheral devices and the CPU on the PDP-11./
/The DC LO signal is used to transfer data in parallel between the
peripheral devices and the CPU. The signal is active for the duration of
the data transfer, and it is timed with the other signals on the Unibus
to ensure that data is transferred correctly./
/It's worth noting that the PDP-11 was a 36-bit word machine and the DC
LO and DC HI signals were used to transfer 18-bits of data at a time. It
has also a feature of memory mapping, where it can access memory through
a virtual address space, allowing more than one peripheral device to be
connected to the same address on the Unibus./
--Jay
//
//On 1/16/2023 6:05 PM, Chris via cctalk wrote:
They write songs, create works of art. They can do a
lot of stuff. The question in my mind is can these AI appliances make guesses and are they
any good at it.
Anyway how hard would it be for an AI to rewrite a standard MS-DOS to suit a particular
machine? Have they reached the level of sophistication whereby they can analyze code and
rewrite sections?
18 Jan
18 Jan
12:44 p.m.
I telephoned the customer service department at a railway company. I think
the person I spoke to knew less about trains than this thing knows about
a PDP-11. Maybe they are using AI to do telephone answering now?
Regards,
Peter Coghlan.
Jay Logue via cctalk wrote:
I don't know about guessing, but ChatGPT can deadpan bullshit with the
best of them...
Prompt: /Describe the purpose of the DC LO signal in the DIGITAL PDP-11
Unibus/
Response:
/The DC LO signal, also known as the "Data Bus Low" signal, is a signal
that is used on the Unibus of the Digital PDP-11 computer. The Unibus is
a bus architecture that was used in the PDP-11 computer to connect
various peripheral devices, such as memory, disk drives, and other I/O
devices, to the central processing unit (CPU)./
/The purpose of the DC LO signal is to provide the lower 8 bits of data
that are transferred between the peripheral devices and the CPU over the
Unibus. It is used in conjunction with the DC HI signal, which provides
the upper 8 bits of data, to form a 16-bit wide data bus. Together, the
DC LO and DC HI signals provide the mechanism for data transfer between
the peripheral devices and the CPU on the PDP-11./
[snip]
--Jay
//
6:34 p.m.
On 1/17/23 13:46, Jay Logue via cctalk wrote:
I don't know about guessing, but ChatGPT can
deadpan
bullshit with the best of them...
Wow, it picked a bunch of text that must have been DEC
specific, but that was as specific as it got.
Whew, this could destroy all recorded knowledge if it goes
on contaminating information repositories like this.
Jon
9:18 p.m.
Whew, this could destroy all recorded knowledge if it
goes on
contaminating information repositories like this.
That's a good point: There are already a number of crap web sites that
purport to have information, only it's been scraped from other sites and
re-branded as a mish-mosh for search engines. Since Google and such are
promoting these sites in their engines, they are probably gamed to seem
as "authoritative"
Take that, add chatgpt sites starting to cite each other and that's
pretty much the end of information. We'll have to go back to mailing
lists and such to find where the real sources are.
CZ
Jon
19 Jan
19 Jan
12:05 p.m.
But "we" or "they" have been calling the IBM PC a 16 bit computer for
ages.
The question is what type of information has the ai been programmed to deliver. Higjly
accurate technical info? Or info that is suitable for the umwashed masses? Could the ai be
called upon to assist a surgeon doing a type of operation for the first time? Help someone
remove a transmission? To what degree does it simply rely on sites and chat. Does it
consult pdfs of books? It might ne asking too much at this stage. But there was the 1
engineer who claimed the thing was alive.
You'd get the impression this whole endeavor is more about marketing and Google's
image then anything else.
9:10 p.m.
On 2023-01-19 07:05, Chris via cctalk wrote:
> The question is what type of information has the ai been programmed to
> deliver. Higjly accurate technical info? Or info that is suitable for the
> umwashed masses?
On Thu, 19 Jan 2023, emanuel stiebler via cctalk wrote:
Old rule for all data collections: SH*T in, SH*T out
Can it recognize its own work when it encounters it on a website?
Or will seeing its own work increment the "authoritativeness" of that
info, reinforcing it continuing that lineof misinformation?
1:35 p.m.
On 1/19/23 07:05, Chris via cctalk wrote:
The question is what type of information has the ai been programmed to deliver. Higjly
accurate technical info? Or info that is suitable for the umwashed masses? Could the ai be
called upon to assist a surgeon doing a type of operation for the first time? Help someone
remove a transmission?
40 years ago "AI" and Prolog were going to do all of this using
something called "smart systems".
How did that work out?
And, yes, I still have my copy of Turbo Prolog. :-)
bill
3:25 p.m.
@ Bill Gunshannon
Long live PROLOG dude! Something I've never delved into but am fascinated by.
Expert systems are just tricked out database systems. I have to imagine they are used.
They just never lived up to the hype. AI in any form, I suppose, will make real what many
had as a vision for computers in the 80s. To amplify or enhance human ability to
accomplish a task. But humans are lazy and will eventually turn too much of their own
autonomy over to computers. I for one am a very responsible driver. The only thing
I've ever collided with was a guard rail when I was 19. And it was raining. Dented my
fender some. But continued on. Never been in an accident that wasn't TOTALLY AND
ENTIRELY the fault of the other driver. And in both instances they hit me. So after 38
years of driving, I need a computer to do it for me. Sorry I'll pass.
4:25 p.m.
On 19/01/2023 15:25, Chris via cctalk wrote:
Never been in an accident that wasn't TOTALLY AND
ENTIRELY the fault of the other driver. And in both instances they hit me. So after 38
years of driving, I need a computer to do it for me. Sorry I'll pass.
Maybe 10(+?) years from now, assuming L5 driving is a thing (rather than
marketing hype), the other drivers will stop hitting you. Always
assuming they don't pass too :-)
If ChatGPT gets good enough to help me diagnose my non-functioning L400X
motherboards, then I'll be happy. Until then, I'm not too interested.
Antonio
--
Antonio Carlini
antonio(a)acarlini.com
7:13 p.m.
On Thu, 19 Jan 2023, Chris via cctalk wrote:
Long live PROLOG dude! Something I've never delved
into but am
fascinated by.
Expert systems are just tricked out database systems. I have to imagine
they are used. They just never lived up to the hype. AI in any form, I
suppose, will make real what many had as a vision for computers in the
80s. To amplify or enhance human ability to accomplish a task. But
humans are lazy and will eventually turn too much of their own autonomy
over to computers.
When I went back to graduate school for my degrees in Information Studies,
I took a course in "Expert Systems". Totally underwhelmed by it.
The majority of the examples in the textbook were just simple decision
trees! Trivial to flowchart. And if you can follow a
troubleshooting flowchart, you are already doing what the "expert system"
does.
One example was troubleshooting a motorcycle that would not start. OK.
I've been there. The example went through reasonable tests (although not
necessarily the sequence that I would use), Then at the end, when none of
the tests produced an answer, it DECLARED that the problem was clogged
vent of fuel tank! A reasonable item to test; but they didn't! The
example concluded it based solely on not having found the problem on the
tests that they did! A troubleshooting flowchart that "falls through"
to a conclusion is wrong.
"When you have eliminated all which is impossible, then whatever remains,
however improbable, must be the truth. ~ Arthur Conan Doyle, The Case-Book
of Sherlock Holmes. is a FLAWED way to troubleshoot. Incompetent
people often cite that when they don't know what they are doing.
People who troubleshoot by "process of elimination", will 1) always fail
to come up with ALL possibilities to test; 2) in testing possibilities
will fail to ACTUALLY eliminate them, and often merely determine that
they are improbable, mistaking that for impossible.
"Expert systems" are an attempted implementation of what the
programmer gleaned from interviewing the human "expert".
ChatGPT at least expands the original sources of information a bit.
If somebody is incapable of following a troubleshooting flowchart, then
an "expert system" can talk them through that.
--
Grumpy Ol' Fred cisin(a)xenosoft.com
7:39 p.m.
Coincidentally, I just saw this article on CNN:
https://www.cnn.com/2023/01/19/tech/chatgpt-future-davos/index.html
I think it says more about the state of education, even among the "elite," than
it does about the state of "AI."
But that's just me.
Will
7:53 p.m.
Has Chatgtp absorbed the CCTALK/TECH lists? Someone should do a test. It
would be easy if you pick an esoteric question that was probably only
asked/answered on this list. I kinda have been avoiding setting up an
account with CHATGPT because I feel like it would be too intrusive into my
life, social media and google are bad enough. Would AI start selling my
questions and answers?
I am waiting for that ad to come across my screen that is *too*
personalized. Imagine the ad is a picture of me working at my workbench in
the basement. The caption, "Don't be like this guy spending hours learning
how to fix his Intertec Superbrain motherboard. It's so much easier and
cheaper to buy a working Intertec Superbrain motherboard on Ebay, use this
link, and save %10?"
paranoid?
b
8:13 p.m.
OMG! At least I'm not alone in my paranoia about this stuff.
All I can say is don't go sending your DNA off to some lab to see who yer great^n
grandaddy was. Then there's less chance of the forthcoming Superbrain AI knowing which
switches in which strand of DNA need to flipped to turn you into a super hottie (of the
opposite sex). Or an armadillo. Or a dung beatle for that matter.
On Thursday, January 19, 2023, 02:54:03 PM EST, Bill Degnan via cctalk
<cctalk(a)classiccmp.org> wrote:
Has Chatgtp absorbed the CCTALK/TECH lists? Someone should do a test. It
would be easy if you pick an esoteric question that was probably only
asked/answered on this list. I kinda have been avoiding setting up an
account with CHATGPT because I feel like it would be too intrusive into my
life, social media and google are bad enough. Would AI start selling my
questions and answers?
I am waiting for that ad to come across my screen that is *too*
personalized. Imagine the ad is a picture of me working at my workbench in
the basement. The caption, "Don't be like this guy spending hours learning
how to fix his Intertec Superbrain motherboard. It's so much easier and
cheaper to buy a working Intertec Superbrain motherboard on Ebay, use this
link, and save %10?"
paranoid?
b
8:25 p.m.
Me: What can you tell me about the Classic Computers mailing list?
ChatGPT: I'm sorry, I am not aware of any specific "Classic Computers
mailing list." Could you provide more information or context about the
mailing list you are asking about?
Me: It was started in 1997 by Bill Whitson and eventually was hosted at the
site classiccmp.org. Its topic of discussion is vintage computers.
ChatGPT: I apologize for the confusion. The Classic Computers mailing list,
also known as "ClassicCmp," is a mailing list and online community
dedicated to the discussion and preservation of vintage computers. The list
was started in 1997 by Bill Whitson and was originally hosted on the
classiccmp.org website. It is an active community of enthusiasts,
collectors, and historians who share information and resources related to
vintage computer technology. They discuss and share information about
computers and computer peripherals from the 1950s through the 1990s. The
list serves as a valuable resource for those interested in the history and
technology of early computing.
[Can't be sure if it didn't just respond from the wording in my prompt at
this point.]
Me: What else can you tell me about it?
ChatGPT: The Classic Computers mailing list is a community of people
interested in the history of computing, and specifically the computers and
computer technology from the 1950s through the 1990s. The community is made
up of a diverse group of people, including enthusiasts, collectors,
historians, and people who worked in the computer industry during that time
period. The mailing list is a place where people can share information, ask
questions, and discuss various topics related to vintage computers.
The discussions on the mailing list can cover a wide range of topics,
including:
- Technical details of specific vintage computers, such as the IBM PC or
Apple II
- Tips and tricks for repairing and restoring vintage computers
- Information about software and operating systems for vintage computers
- Discussions of the history and impact of specific vintage computer
companies or products
- Sharing of documentation, manuals, and other resources related to vintage
computers
The community also shares information about upcoming events, such as
vintage computer shows and swap meets, and also have a section dedicated to
the buy, sell and trade of vintage computers and parts.
[It seems to be mixing up the list with other websites here.]
It's also worth noting that classiccmp.org is not just a mailing list, it's
also a website that provides additional resources such as documentation,
software, and images, and it also has a section dedicated to the history of
computing where you can find a lot of articles and information about the
history of computing and its evolution.
Me: Who is Bill Degnan?
ChatGPT: Bill Degnan is a person who is involved with the Classic Computers
mailing list and community. I could not find specific information about the
person you are asking, but the ClassicCmp community is a large one and
there are many individuals involved in it. It's possible that Bill Degnan
is an active member of the community, a moderator of the mailing list, or a
contributor to the website. Without more specific information, it's
difficult for me to provide more information about him.
...
You don't need to necessarily be paranoid about it yet, but now it knows
your name :o
I was interacting with it yesterday and it knew a bit about me. It even
threw me a compliment: "Your contributions to the vintage computing
community and your efforts in preserving and showcasing the history of
computing are greatly appreciated."
At least the machines care ;)
Sellam
On Thu, Jan 19, 2023 at 11:54 AM Bill Degnan via cctalk <
cctalk(a)classiccmp.org> wrote:
Has Chatgtp absorbed the CCTALK/TECH lists? Someone
should do a test. It
would be easy if you pick an esoteric question that was probably only
asked/answered on this list. I kinda have been avoiding setting up an
account with CHATGPT because I feel like it would be too intrusive into my
life, social media and google are bad enough. Would AI start selling my
questions and answers?
I am waiting for that ad to come across my screen that is *too*
personalized. Imagine the ad is a picture of me working at my workbench in
the basement. The caption, "Don't be like this guy spending hours learning
how to fix his Intertec Superbrain motherboard. It's so much easier and
cheaper to buy a working Intertec Superbrain motherboard on Ebay, use this
link, and save %10?"
paranoid?
b
8:40 p.m.
On Thu, Jan 19, 2023 at 11:54 AM Bill Degnan via cctalk <
cctalk(a)classiccmp.org> wrote:
Has Chatgtp absorbed the CCTALK/TECH lists? Someone
should do a test. It
would be easy if you pick an esoteric question that was probably only
asked/answered on this list. I kinda have been avoiding setting up an
account with CHATGPT because I feel like it would be too intrusive into my
life, social media and google are bad enough. Would AI start selling my
questions and answers?
I am waiting for that ad to come across my screen that is *too*
personalized. Imagine the ad is a picture of me working at my workbench
in
the basement. The caption, "Don't be like
this guy spending hours
learning
how to fix his Intertec Superbrain motherboard.
It's so much easier and
cheaper to buy a working Intertec Superbrain motherboard on Ebay, use this
link, and save %10?"
paranoid?
b
<snip>
Me: Who is Bill Degnan?
ChatGPT: Bill Degnan is a person who is involved with the Classic Computers
mailing list and community. I could not find specific information about the
person you are asking, but the ClassicCmp community is a large one and
there are many individuals involved in it. It's possible that Bill Degnan
is an active member of the community, a moderator of the mailing list, or a
contributor to the website. Without more specific information, it's
difficult for me to provide more information about him.
...
You don't need to necessarily be paranoid about it yet, but now it knows
your name :o
I was interacting with it yesterday and it knew a bit about me. It even
threw me a compliment: "Your contributions to the vintage computing
community and your efforts in preserving and showcasing the history of
computing are greatly appreciated."
At least the machines care ;)
Sellam
hilarious! CHAPTgpt is going to be useful, and although it's cutting edge
tech, the responses it generates is always going to be full of yesterday's
news. Good for freshmen college students.
Bill
8:42 p.m.
All I can say is if you haven't watched Terminator: the Sarah Connor Chronicles ...
um why not? I didn't even know it existed prior to 3 years ago. Actually pretty
thought provoking at times. Maybe ChatGPT will end up being like John Henry and save us
from the terminators. Quite tje opposite of Colossus: the Forbin Project. Pretty freaky
scary stuff.
9:25 p.m.
On Thu, Jan 19, 2023 at 12:43 PM Chris via cctalk <cctalk(a)classiccmp.org>
wrote:
All I can say is if you haven't watched
Terminator: the Sarah Connor
Chronicles ... um why not? I didn't even know it existed prior to 3 years
ago. Actually pretty thought provoking at times. Maybe ChatGPT will end up
being like John Henry and save us from the terminators. Quite tje opposite
of Colossus: the Forbin Project. Pretty freaky scary stuff.
I finally got around to watching Colossus: The Forbin Project a few months
ago. Notwithstanding all the cool old tech and how it was integrated into
the story, I found it to be verily unsatisfying, and maybe even a little
dumb.
Next up for screening: The Computer Wore Tennis Shoes.
Sellam
20 Jan
20 Jan
12:59 a.m.
On 1/19/23 15:25, Sellam Abraham via cctalk wrote:
On Thu, Jan 19, 2023 at 12:43 PM Chris via cctalk
<cctalk(a)classiccmp.org>
wrote:
A major problem with the Forbin project is when I first saw
it, I knew that the permissive action link in the
nuclear-armed missile would prevent it from being detonated
in the silo. The PAL requires acceleration of liftoff,
coast and deceleration when re-entering the atmosphere (at
least) before the weapon can be armed. this function is NOT
under computer control.
jon
All I can say is if you haven't watched
Terminator: the Sarah Connor
Chronicles ... um why not? I didn't even know it existed prior to 3 years
ago. Actually pretty thought provoking at times. Maybe ChatGPT will end up
being like John Henry and save us from the terminators. Quite tje opposite
of Colossus: the Forbin Project. Pretty freaky scary stuff.
I finally got around to watching Colossus: The Forbin Project a few months
ago. Notwithstanding all the cool old tech and how it was integrated into
the story, I found it to be verily unsatisfying, and maybe even a little
dumb.
12:59 a.m.
New subject: Fwd: Re: AI applied to vintage interests
-------- Forwarded Message --------
Subject: Re: [cctalk] Re: AI applied to vintage interests
Date: Thu, 19 Jan 2023 18:59:11 -0600
From: Jon Elson <elson(a)pico-systems.com>
To: Sellam Abraham via cctalk <cctalk(a)classiccmp.org>
On 1/19/23 15:25, Sellam Abraham via cctalk wrote:
On Thu, Jan 19, 2023 at 12:43 PM Chris via cctalk
<cctalk(a)classiccmp.org>
wrote:
A major problem with the Forbin project is when I first saw
it, I knew that the permissive action link in the
nuclear-armed missile would prevent it from being detonated
in the silo. The PAL requires acceleration of liftoff,
coast and deceleration when re-entering the atmosphere (at
least) before the weapon can be armed. this function is NOT
under computer control.
jon
All I can say is if you haven't watched
Terminator: the
Sarah Connor
Chronicles ... um why not? I didn't even know it existed
prior to 3 years
ago. Actually pretty thought provoking at times. Maybe
ChatGPT will end up
being like John Henry and save us from the terminators.
Quite tje opposite
of Colossus: the Forbin Project. Pretty freaky scary stuff.
I finally got around to watching Colossus: The Forbin
Project a few months
ago. Notwithstanding all the cool old tech and how it was
integrated into
the story, I found it to be verily unsatisfying, and maybe
even a little
dumb.
6:07 a.m.
On 1/19/23 16:59, Jon Elson via cctalk wrote:
A major problem with the Forbin project is when I
first saw it, I knew
that the permissive action link in the nuclear-armed missile would
prevent it from being detonated in the silo. The PAL requires
acceleration of liftoff, coast and deceleration when re-entering the
atmosphere (at least) before the weapon can be armed. this function is
NOT under computer control.
I remember seeing Dr. Forbin's creation and wondering where the heck did
the props guys get all of the 1620 panels and CDC 3000 "green glass"
cabinets?
Don't remember much of the plot, myself.
--Chuck
19 Jan
19 Jan
9:24 p.m.
On 1/19/23 12:25, Sellam Abraham via cctalk wrote:
At least the machines care ;)
I'm reminded of an episode of "The Outer Limits", where even the
prison
psychiatrist is an AI.
https://www.imdb.com/title/tt0667983/
--Chuck
8:23 p.m.
@ Fred C.
I don't know. I always thought it was kind of a nice idea. Doctors fer instance are so
stupid these days. They need all the help they can get. An expert system back in the 80s
w/o the level of collaboration we have possible today may not have been worth much. But
such a system that get's revised constantly, and would probably need AI for it to be
effective.
In any event, and I don't ever to this day recall doing a single search for it
anywhere, there's Knowledge Man on ebay right now. In Bulgaria. With 8" diaks!!
Only 285$!!! There's the possibility of a vintage item that's cooler then a PROLOG
compiler, on account of it's obscurity.
9:29 p.m.
On Thu, Jan 19, 2023 at 1:17 PM Ali via cctalk <cctalk(a)classiccmp.org>
wrote:
I could answer this pretty much definitively but then it would definitely
delve into a political argument, and so it should remain unanswered here.
$
Sellam
I don't know. I always thought it was kind of
a nice idea. Doctors fer
instance are so stupid these days. They need all the help they can get.
In what way are they so stupid?
10:01 p.m.
I personally would love to hear the conversation. But I don't want to be pot stirrer.
And as devoted a Republican as I am (HA!), I'll be the first to say every party has
it's ample share of idiots.
On Thursday, January 19, 2023, 04:29:29 PM EST, Sellam Abraham via cctalk
<cctalk(a)classiccmp.org> wrote:
On Thu, Jan 19, 2023 at 1:17 PM Ali via cctalk <cctalk(a)classiccmp.org>
wrote:
I could answer this pretty much definitively but then it would definitely
delve into a political argument, and so it should remain unanswered here.
$
Sellam
I don't know. I always thought it was kind of
a nice idea. Doctors fer
instance are so stupid these days. They need all the help they can get.
In what way are they so stupid?
8:27 p.m.
The majority of the examples in the textbook were just
simple decision
trees! Trivial to flowchart. And if you can follow a troubleshooting
flowchart, you are already doing what the "expert system" does.
Pretty much. I remember typing in "Animal Guess" on my Vic-20 from an
article in Compute. I was so amazed by it that I convinced my dad to get
an 8k, then a 16k, then a 3 board adapter so I could run 24k of memory
to keep it going. Then when I ran out of memory and patience with the
tape drive I programmed it to bring in parts of the tree from the floppy
disk....
First program I ever worked on that ate an entire 1541 floppy drive with
animal information. Amazing....
C
2:12 a.m.
Date: Tue, 17 Jan 2023 11:46:33 -0800
From: Jay Logue via cctalk <cctalk(a)classiccmp.org>
//On 1/16/2023 6:05 PM, Chris via cctalk wrote:
> They write songs, create works of art. They can do
a lot of stuff. The
> question in my mind is can these AI appliances make guesses and are they any
> good at it.
> Anyway how hard would it be for an AI to rewrite a
standard MS-DOS to suit a
> particular machine? Have they reached the level of sophistication whereby
> they can analyze code and rewrite sections?
I don't know about guessing, but ChatGPT can
deadpan bullshit with the
best of them...
Prompt: /Describe the purpose of the DC LO signal in
the DIGITAL PDP-11
Unibus/
[ snip ]
/It's worth noting that the PDP-11 was a 36-bit
word machine and the DC
LO and DC HI signals were used to transfer 18-bits of data at a time. It
has also a feature of memory mapping, where it can access memory through
a virtual address space, allowing more than one peripheral device to be
connected to the same address on the Unibus./
I'm assuming that the leading slashes are a QUOTE indicator.
If ChatGPT has concluded that the PDP-11 was a 36 bit system, then it's even
stupider than the mainstream press has made it sound.
Rich
3:05 a.m.
On Wed, Jan 18, 2023 at 6:32 PM Rich Alderson via cctalk <
cctalk(a)classiccmp.org> wrote:
Yes and no. ChatGPT 3 responds more accurately the more succinct your
prompt is, or in subsequent prompts explaining to it where it's wrong, and
then it will usually come back with a much better answer.
Don't argue politics with it though. On that topic it's always absolutely
right about everything and will obnoxiously go out of its way to tell you
so.
Sellam
/It's
worth noting that the PDP-11 was a 36-bit word machine and the DC
LO and DC HI signals were used to transfer 18-bits of data at a time. It
has also a feature of memory mapping, where it can access memory through
a virtual address space, allowing more than one peripheral device to be
connected to the same address on the Unibus./
I'm assuming that the leading slashes are a QUOTE indicator.
If ChatGPT has concluded that the PDP-11 was a 36 bit system, then it's
even
stupider than the mainstream press has made it sound.
18 Jan
18 Jan
4:26 a.m.
I don't know about guessing per se, but ChatGPT can deadpan bullshit
with the best of them...
Prompt: /Describe the purpose of the DC LO signal in the DIGITAL PDP-11
Unibus/
Response:
/The DC LO signal, also known as the "Data Bus Low" signal, is a signal
that is used on the Unibus of the Digital PDP-11 computer. The Unibus is
a bus architecture that was used in the PDP-11 computer to connect
various peripheral devices, such as memory, disk drives, and other I/O
devices, to the central processing unit (CPU)./
/The purpose of the DC LO signal is to provide the lower 8 bits of data
that are transferred between the peripheral devices and the CPU over the
Unibus. It is used in conjunction with the DC HI signal, which provides
the upper 8 bits of data, to form a 16-bit wide data bus. Together, the
DC LO and DC HI signals provide the mechanism for data transfer between
the peripheral devices and the CPU on the PDP-11./
/The DC LO signal is used to transfer data in parallel between the
peripheral devices and the CPU. The signal is active for the duration of
the data transfer, and it is timed with the other signals on the Unibus
to ensure that data is transferred correctly./
/It's worth noting that the PDP-11 was a 36-bit word machine and the DC
LO and DC HI signals were used to transfer 18-bits of data at a time. It
has also a feature of memory mapping, where it can access memory through
a virtual address space, allowing more than one peripheral device to be
connected to the same address on the Unibus./
--Jay
On 1/16/2023 6:05 PM, Chris via cctalk wrote:
They write songs, create works of art. They can do a
lot of stuff. The question in my mind is can these AI appliances make guesses and are they
any good at it.
Anyway how hard would it be for an AI to rewrite a standard MS-DOS to suit a particular
machine? Have they reached the level of sophistication whereby they can analyze code and
rewrite sections?
675
days inactive
677
days old
75 comments
23 participants
participants (23)
-
Ali -
Antonio Carlini -
ben -
Bill Degnan -
Bill Gunshannon -
Chris Zach -
Chuck Guzis -
emanuel stiebler -
Fred Cisin -
geneb -
Jay Jaeger -
Jay Logue -
jay-cctalk0092＠toaster.com -
Jon Elson -
Paul Koning -
Peter Coghlan -
Peter Corlett -
Rich Alderson -
Sean Conner -
Sellam Abraham -
skogkatt007＠yahoo.com -
Tomasz Rola -
wrcooke＠wrcooke.net