Fwd: Re: solar cell mount math fail

RW
Raymond West
Thu, May 25, 2023 11:36 AM

In order to use it,you need to know that it is not just a calculator. If
you grew up with slide rules, then the answer is probably good enough in
that instance. In my simple testing, it seems to only deal to the first
two dp's. e.g 0.69/0.26. (and until I asked for higher resolution, it
gave the correct answer to 2 dp. )

I'm hoping, that in later iterations, it will be able to test it's own
answers, but as is, it generally eventually gets to my desired result.

On 24/05/2023 18:26, Jordan Brown wrote:

On 5/23/2023 4:51 PM, Jordan Brown wrote:

Also, I don't know whether ChatGPT will ever get basic math wrong,
but it does get facts about the real world wrong
https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence),
so I would double-check anything that it says.

Rogier Wolff points out privately that ChatGPT does get basic math wrong.

Note that in the article ChatGPT blames its ~20% error on floating
point imprecision in Python.  That's totally bogus.

I asked ChatGPT 3.5 the same question that the writer did, and it got
it closer, but still wrong by about 0.1%.  It said

`````` t ≈ (0.6931 / 0.2624) ≈ 2.639.
``````

but it isn't; it's actually more than 2.641.

what is 0.6931 divided by 0.2624?

<<< 0.6931 divided by 0.2624 is approximately equal to 2.639.

Let's multiply both of them by 10,000.

What is 6931 divided by 2624?

<<< 6931 divided by 2624 is approximately equal to 2.64125.

This is pretty poor.

To unsubscribe send an email todiscuss-leave@lists.openscad.org

In order to use it,you need to know that it is not just a calculator. If you grew up with slide rules, then the answer is probably good enough in that instance. In my simple testing, it seems to only deal to the first two dp's. e.g 0.69/0.26. (and until I asked for higher resolution, it gave the correct answer to 2 dp. ) I'm hoping, that in later iterations, it will be able to test it's own answers, but as is, it generally eventually gets to my desired result. On 24/05/2023 18:26, Jordan Brown wrote: > On 5/23/2023 4:51 PM, Jordan Brown wrote: >> Also, I don't know whether ChatGPT will ever get basic math wrong, >> but it *does* get facts about the real world wrong >> <https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)>, >> so I would double-check anything that it says. > > Rogier Wolff points out privately that ChatGPT does get basic math wrong. > https://medium.datadriveninvestor.com/i-found-a-simple-math-mistake-in-chatgpt-ac83a7994f10 > > Note that in the article ChatGPT blames its ~20% error on floating > point imprecision in Python.  That's totally bogus. > > I asked ChatGPT 3.5 the same question that the writer did, and it got > it closer, but still wrong by about 0.1%.  It said > > t ≈ (0.6931 / 0.2624) ≈ 2.639. > > but it isn't; it's actually more than 2.641. > > >>> what is 0.6931 divided by 0.2624? > <<< 0.6931 divided by 0.2624 is approximately equal to 2.639. > > Let's multiply both of them by 10,000. > > >>> What is 6931 divided by 2624? > <<< 6931 divided by 2624 is approximately equal to 2.64125. > > This is pretty poor. > > _______________________________________________ > OpenSCAD mailing list > To unsubscribe send an email todiscuss-leave@lists.openscad.org
NH
Thu, May 25, 2023 11:43 AM

I use the version built into Bing, which is ChatGPT4 I believe and it is
much better.

0.6931 divided by 0.2624 is approximately 2.6413872
https://bing.com/search?q=0.6931+%2f+0.2624

On Thu, 25 May 2023 at 12:36, Raymond West raywest@raywest.com wrote:

In order to use it,you need to know that it is not just a calculator. If
you grew up with slide rules, then the answer is probably good enough in
that instance. In my simple testing, it seems to only deal to the first two
dp's. e.g 0.69/0.26. (and until I asked for higher resolution, it gave the
correct answer to 2 dp. )

I'm hoping, that in later iterations, it will be able to test it's own
answers, but as is, it generally eventually gets to my desired result.
On 24/05/2023 18:26, Jordan Brown wrote:

On 5/23/2023 4:51 PM, Jordan Brown wrote:

Also, I don't know whether ChatGPT will ever get basic math wrong, but it
does get facts about the real world wrong
https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence),
so I would double-check anything that it says.

Rogier Wolff points out privately that ChatGPT does get basic math wrong.

Note that in the article ChatGPT blames its ~20% error on floating point
imprecision in Python.  That's totally bogus.

I asked ChatGPT 3.5 the same question that the writer did, and it got it
closer, but still wrong by about 0.1%.  It said

t ≈ (0.6931 / 0.2624) ≈ 2.639.

but it isn't; it's actually more than 2.641.

what is 0.6931 divided by 0.2624?

<<< 0.6931 divided by 0.2624 is approximately equal to 2.639.

Let's multiply both of them by 10,000.

What is 6931 divided by 2624?

<<< 6931 divided by 2624 is approximately equal to 2.64125.

This is pretty poor.

To unsubscribe send an email to discuss-leave@lists.openscad.org

To unsubscribe send an email to discuss-leave@lists.openscad.org

I use the version built into Bing, which is ChatGPT4 I believe and it is much better. 0.6931 divided by 0.2624 is approximately *2.6413872* <https://bing.com/search?q=0.6931+%2f+0.2624> On Thu, 25 May 2023 at 12:36, Raymond West <raywest@raywest.com> wrote: > In order to use it,you need to know that it is not just a calculator. If > you grew up with slide rules, then the answer is probably good enough in > that instance. In my simple testing, it seems to only deal to the first two > dp's. e.g 0.69/0.26. (and until I asked for higher resolution, it gave the > correct answer to 2 dp. ) > > I'm hoping, that in later iterations, it will be able to test it's own > answers, but as is, it generally eventually gets to my desired result. > On 24/05/2023 18:26, Jordan Brown wrote: > > On 5/23/2023 4:51 PM, Jordan Brown wrote: > > Also, I don't know whether ChatGPT will ever get basic math wrong, but it > *does* get facts about the real world wrong > <https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)>, > so I would double-check anything that it says. > > > Rogier Wolff points out privately that ChatGPT does get basic math wrong. > > https://medium.datadriveninvestor.com/i-found-a-simple-math-mistake-in-chatgpt-ac83a7994f10 > > Note that in the article ChatGPT blames its ~20% error on floating point > imprecision in Python. That's totally bogus. > > I asked ChatGPT 3.5 the same question that the writer did, and it got it > closer, but still wrong by about 0.1%. It said > > t ≈ (0.6931 / 0.2624) ≈ 2.639. > > but it isn't; it's actually more than 2.641. > > >>> what is 0.6931 divided by 0.2624? > <<< 0.6931 divided by 0.2624 is approximately equal to 2.639. > > Let's multiply both of them by 10,000. > > >>> What is 6931 divided by 2624? > <<< 6931 divided by 2624 is approximately equal to 2.64125. > > This is pretty poor. > > _______________________________________________ > OpenSCAD mailing list > To unsubscribe send an email to discuss-leave@lists.openscad.org > > _______________________________________________ > OpenSCAD mailing list > To unsubscribe send an email to discuss-leave@lists.openscad.org >
RW
Raymond West
Thu, May 25, 2023 6:14 PM

You're dead right there, wrt Bing. I'm getting it to write a pretty
comprehensive batch file, to process an openscad svg and generate a
specific version of g-code. It started off by saying it couldn't parse a
file, but it lied. It is taking a while, as I test each iteration, but
it will get there. The result will have plenty of room for improvement,
but It will work good enough.

On 25/05/2023 12:43, nop head wrote:

I use the version built into Bing, which is ChatGPT4 I believe and it
is much better.

0.6931 divided by 0.2624 is approximately 2.6413872
https://bing.com/search?q=0.6931+%2f+0.2624

On Thu, 25 May 2023 at 12:36, Raymond West raywest@raywest.com wrote:

`````` In order to use it,you need to know that it is not just a
calculator. If you grew up with slide rules, then the answer is
probably good enough in that instance. In my simple testing, it
seems to only deal to the first two dp's. e.g 0.69/0.26. (and
until I asked for higher resolution, it gave the correct answer to
2 dp. )

I'm hoping, that in later iterations, it will be able to test it's
own answers, but as is, it generally eventually gets to my desired
result.

On 24/05/2023 18:26, Jordan Brown wrote:
``````
`````` On 5/23/2023 4:51 PM, Jordan Brown wrote:
``````
`````` Also, I don't know whether ChatGPT will ever get basic math
wrong, but it *does* get facts about the real world wrong
<https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)>,
so I would double-check anything that it says.
``````
`````` Rogier Wolff points out privately that ChatGPT does get basic
math wrong.

Note that in the article ChatGPT blames its ~20% error on
floating point imprecision in Python.  That's totally bogus.

I asked ChatGPT 3.5 the same question that the writer did, and it
got it closer, but still wrong by about 0.1%.  It said

t ≈ (0.6931 / 0.2624) ≈ 2.639.

but it isn't; it's actually more than 2.641.
``````

what is 0.6931 divided by 0.2624?

`````` <<< 0.6931 divided by 0.2624 is approximately equal to 2.639.

Let's multiply both of them by 10,000.
``````

What is 6931 divided by 2624?

`````` <<< 6931 divided by 2624 is approximately equal to 2.64125.

This is pretty poor.

_______________________________________________
To unsubscribe send an email todiscuss-leave@lists.openscad.org
``````
`````` _______________________________________________
To unsubscribe send an email to discuss-leave@lists.openscad.org
``````

To unsubscribe send an email todiscuss-leave@lists.openscad.org

You're dead right there, wrt Bing. I'm getting it to write a pretty comprehensive batch file, to process an openscad svg and generate a specific version of g-code. It started off by saying it couldn't parse a file, but it lied. It is taking a while, as I test each iteration, but it will get there. The result will have plenty of room for improvement, but It will work good enough. On 25/05/2023 12:43, nop head wrote: > I use the version built into Bing, which is ChatGPT4 I believe and it > is much better. > > 0.6931 divided by 0.2624 is approximately *2.6413872* > <https://bing.com/search?q=0.6931+%2f+0.2624> > > On Thu, 25 May 2023 at 12:36, Raymond West <raywest@raywest.com> wrote: > > In order to use it,you need to know that it is not just a > calculator. If you grew up with slide rules, then the answer is > probably good enough in that instance. In my simple testing, it > seems to only deal to the first two dp's. e.g 0.69/0.26. (and > until I asked for higher resolution, it gave the correct answer to > 2 dp. ) > > I'm hoping, that in later iterations, it will be able to test it's > own answers, but as is, it generally eventually gets to my desired > result. > > On 24/05/2023 18:26, Jordan Brown wrote: >> On 5/23/2023 4:51 PM, Jordan Brown wrote: >>> Also, I don't know whether ChatGPT will ever get basic math >>> wrong, but it *does* get facts about the real world wrong >>> <https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)>, >>> so I would double-check anything that it says. >> >> Rogier Wolff points out privately that ChatGPT does get basic >> math wrong. >> https://medium.datadriveninvestor.com/i-found-a-simple-math-mistake-in-chatgpt-ac83a7994f10 >> >> Note that in the article ChatGPT blames its ~20% error on >> floating point imprecision in Python.  That's totally bogus. >> >> I asked ChatGPT 3.5 the same question that the writer did, and it >> got it closer, but still wrong by about 0.1%.  It said >> >> t ≈ (0.6931 / 0.2624) ≈ 2.639. >> >> but it isn't; it's actually more than 2.641. >> >> >>> what is 0.6931 divided by 0.2624? >> <<< 0.6931 divided by 0.2624 is approximately equal to 2.639. >> >> Let's multiply both of them by 10,000. >> >> >>> What is 6931 divided by 2624? >> <<< 6931 divided by 2624 is approximately equal to 2.64125. >> >> This is pretty poor. >> >> _______________________________________________ >> OpenSCAD mailing list >> To unsubscribe send an email todiscuss-leave@lists.openscad.org > _______________________________________________ > OpenSCAD mailing list > To unsubscribe send an email to discuss-leave@lists.openscad.org > > > _______________________________________________ > OpenSCAD mailing list > To unsubscribe send an email todiscuss-leave@lists.openscad.org
RW
Ray West
Sun, May 28, 2023 1:09 PM

I thought there may be some interest in how it was proceeding, following

On 25/05/2023 19:14, Raymond West wrote:

You're dead right there, wrt Bing. I'm getting it to write a pretty
comprehensive batch file, to process an openscad svg and generate a
specific version of g-code. It started off by saying it couldn't parse
a file, but it lied. It is taking a while, as I test each iteration,
but it will get there. The result will have plenty of room for
improvement, but It will work good enough.

On 25/05/2023 12:43, nop head wrote:

I use the version built into Bing, which is ChatGPT4 I believe and it
is much better.

So far, managed to generate code to batch process an scad file, doing
various sections, and combining the SVG files. It is struggling with
converting the SVG to  g code.

I've patiently gone through 33 iterations of trying the batch file
script generated by the AI (within Bing), and feeding back the errors or
results. It eventually seems to get into a loop, repeating previous guesses.

This is the input file  (just one line)

M 10,-10 L 60,-10 L 60,-60 L 10,-60 z

The output that I want which should be

G01 Z-5
G01 X10 Y10
G01 X60 Y10
G01 X60 Y60
G01 X10 Y60
G00 Z2

I think the closest it has got is the following batch file

@echo off
setlocal enabledelayedexpansion

set "inputFile=input.txt"
set "outputFile=output.txt"

if exist "%outputFile%" del "%outputFile%"

for /f "usebackq delims=" %%a in ("%inputFile%") do (
set "line=%%a"
set "firstM=1"
for %%b in (!line!) do (
if "%%b" == "M" (
if "!firstM!" == "1" (
echo G01 Z-5 >> "%outputFile%"
set "firstM=0"
)
set coord=
) else if "%%b" == "L" (
echo !coord! >> "%outputFile%"
set coord=
) else if "%%b" == "z" (
echo !coord! >> "%outputFile%"
echo G00 Z2 >> "%outputFile%"
) else (
for /f "tokens=1,2 delims=," %%x in ("%%b") do (
set y=%%y
set /a y=-1*!y!
if not defined coord (
set coord=G01 X%%x Y!y!
)
)
)
)
if defined coord echo !coord! >> "%outputFile%"
)

endlocal

it does not show the Y values, although in earlier attempts it did, but
produced other errors.

I thought there may be some interest in how it was proceeding, following on from a previous thread. On 25/05/2023 19:14, Raymond West wrote: > > You're dead right there, wrt Bing. I'm getting it to write a pretty > comprehensive batch file, to process an openscad svg and generate a > specific version of g-code. It started off by saying it couldn't parse > a file, but it lied. It is taking a while, as I test each iteration, > but it will get there. The result will have plenty of room for > improvement, but It will work good enough. > > On 25/05/2023 12:43, nop head wrote: >> I use the version built into Bing, which is ChatGPT4 I believe and it >> is much better. >> >> So far, managed to generate code to batch process an scad file, doing various sections, and combining the SVG files. It is struggling with converting the SVG to  g code. I've patiently gone through 33 iterations of trying the batch file script generated by the AI (within Bing), and feeding back the errors or results. It eventually seems to get into a loop, repeating previous guesses. This is the input file  (just one line) M 10,-10 L 60,-10 L 60,-60 L 10,-60 z The output that I want which should be G01 Z-5 G01 X10 Y10 G01 X60 Y10 G01 X60 Y60 G01 X10 Y60 G00 Z2 I think the closest it has got is the following batch file @echo off setlocal enabledelayedexpansion set "inputFile=input.txt" set "outputFile=output.txt" if exist "%outputFile%" del "%outputFile%" for /f "usebackq delims=" %%a in ("%inputFile%") do (     set "line=%%a"     set "firstM=1"     for %%b in (!line!) do (         if "%%b" == "M" (             if "!firstM!" == "1" (                 echo G01 Z-5 >> "%outputFile%"                 set "firstM=0"             )             set coord=         ) else if "%%b" == "L" (             echo !coord! >> "%outputFile%"             set coord=         ) else if "%%b" == "z" (             echo !coord! >> "%outputFile%"             echo G00 Z2 >> "%outputFile%"         ) else (             for /f "tokens=1,2 delims=," %%x in ("%%b") do (                 set y=%%y                 set /a y=-1*!y!                 if not defined coord (                     set coord=G01 X%%x Y!y!                 )             )         )     )     if defined coord echo !coord! >> "%outputFile%" ) endlocal it does not show the Y values, although in earlier attempts it did, but produced other errors.
WF
Sun, May 28, 2023 2:07 PM

Neat that you're making progress on this!
I approached it from a rather different perspective and instead used RapCAD for G-code generation:

I've been documenting it at:

and have been using it inside OpenSCAD Graph Editor:

William

-----Original Message-----
From: Ray West raywest@raywest.com
Sent: Sun, May 28, 2023 9:09 am
Subject: [OpenSCAD] OT openai/batch file/svg parse/gcode generate

I thought there may be some interest in how it was proceeding, following

On 25/05/2023 19:14, Raymond West wrote:

You're dead right there, wrt Bing. I'm getting it to write a pretty
comprehensive batch file, to process an openscad svg and generate a
specific version of g-code. It started off by saying it couldn't parse
a file, but it lied. It is taking a while, as I test each iteration,
but it will get there. The result will have plenty of room for
improvement, but It will work good enough.

On 25/05/2023 12:43, nop head wrote:

I use the version built into Bing, which is ChatGPT4 I believe and it
is much better.

So far, managed to generate code to batch process an scad file, doing
various sections, and combining the SVG files. It is struggling with
converting the SVG to  g code.

I've patiently gone through 33 iterations of trying the batch file
script generated by the AI (within Bing), and feeding back the errors or
results. It eventually seems to get into a loop, repeating previous guesses.

This is the input file  (just one line)

M 10,-10 L 60,-10 L 60,-60 L 10,-60 z

The output that I want which should be

G01 Z-5
G01 X10 Y10
G01 X60 Y10
G01 X60 Y60
G01 X10 Y60
G00 Z2

I think the closest it has got is the following batch file

@echo off
setlocal enabledelayedexpansion

set "inputFile=input.txt"
set "outputFile=output.txt"

if exist "%outputFile%" del "%outputFile%"

for /f "usebackq delims=" %%a in ("%inputFile%") do (
set "line=%%a"
set "firstM=1"
for %%b in (!line!) do (
if "%%b" == "M" (
if "!firstM!" == "1" (
echo G01 Z-5 >> "%outputFile%"
set "firstM=0"
)
set coord=
) else if "%%b" == "L" (
echo !coord! >> "%outputFile%"
set coord=
) else if "%%b" == "z" (
echo !coord! >> "%outputFile%"
echo G00 Z2 >> "%outputFile%"
) else (
for /f "tokens=1,2 delims=," %%x in ("%%b") do (
set y=%%y
set /a y=-1*!y!
if not defined coord (
set coord=G01 X%%x Y!y!
)
)
)
)
if defined coord echo !coord! >> "%outputFile%"
)

endlocal

it does not show the Y values, although in earlier attempts it did, but
produced other errors.