Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > alt.comp.os.windows-11 > #35980 > unrolled thread

Adobe is panicking

Back to article view | Back to alt.comp.os.windows-11

Contents

  Adobe is panicking MegaNZ <MegaNZ@newsgroups.com> - 2026-10-09 00:29 +0100
    Re: Adobe is panicking Paul <nospam@needed.invalid> - 2026-10-09 01:59 -0400
      Re: Adobe is panicking Steve Hayes <hayesstw@telkomsa.net> - 2026-10-10 06:22 +0200
      Re: Adobe is panicking WolfFan <akwolffan@zoho.com> - 2026-10-11 10:13 -0400
    Re: Adobe is panicking "....winston" <winstonmvp@gmail.com> - 2026-10-09 03:21 -0400
      Re: Adobe is panicking dillinger <dillinger@not.invalid> - 2026-10-09 20:02 +0200
        Re: Adobe is panicking "....winston" <winstonmvp@gmail.com> - 2026-10-09 20:53 -0400
          Re: Adobe is panicking Paul <nospam@needed.invalid> - 2026-10-09 21:46 -0400
            Re: Adobe is panicking "....winston" <winstonmvp@gmail.com> - 2026-10-13 13:49 -0400
              Re: Adobe is panicking Paul <nospam@needed.invalid> - 2026-10-13 16:38 -0400
                Re: Adobe is panicking "....winston" <winstonmvp@gmail.com> - 2026-10-15 11:35 -0400
                  Re: Adobe is panicking "Jeff Gaines" <jgnewsid@outlook.com> - 2026-10-15 20:50 +0000
                    Re: Adobe is panicking "....winston" <winstonmvp@gmail.com> - 2026-10-16 02:04 -0400
                      [OT[: "24/7/365" (was: Re: Adobe is panicking) "J. P. Gilliver" <G6JPG@255soft.uk> - 2026-10-16 08:16 +0100
              Re: Adobe is panicking Lars Poulsen <lars@beagle-ears.com> - 2026-10-13 18:52 -0700
    Re: Adobe is panicking Brian Gregory <void-invalid-dead-dontuse@email.invalid> - 2026-10-11 19:21 +0100
      Re: Adobe is panicking Andy Burns <usenet@andyburns.uk> - 2026-10-11 22:39 +0100
        Re: Adobe is panicking Paul <nospam@needed.invalid> - 2026-10-12 00:00 -0400
        Re: Adobe is panicking Brian Gregory <void-invalid-dead-dontuse@email.invalid> - 2026-10-12 14:06 +0100
          Re: Adobe is panicking Paul <nospam@needed.invalid> - 2026-10-12 18:24 -0400
            Re: Adobe is panicking Brian Gregory <void-invalid-dead-dontuse@email.invalid> - 2026-10-15 23:15 +0100

Page 1 of 2  [1] 2  Next page →

#35980 — Adobe is panicking

According to YouTube videos, Adobe must be having a nightmare, not to 
mention sleepless nights!

<https://youtu.be/doHqV2t26KU>
<https://www.youtube.com/results?search_query=ArtCraft>

Just 48 hours after Vibe released his products, Adobe called a special 
board meeting and requested an urgent meeting with the Trump 
administration to discuss ways to stop AI reverse-engineering Adobe 
products.

It's a worrying time for the greedy corporation that made the 
subscription model mandatory in the software industry — we saw HP's CEO 
leave the firm for PayPal when HP customers started switching to Ecotank 
printers, which don't require a subscription, and we might see the same 
thing happen at Adobe.

[toc] | [next] | [standalone]

#35989

On Thu, 10/8/2026 7:29 PM, MegaNZ wrote:
> According to YouTube videos, Adobe must be having a nightmare, not to 
> mention sleepless nights!
> 
> <https://youtu.be/doHqV2t26KU>
> <https://www.youtube.com/results?search_query=ArtCraft>
> 
> Just 48 hours after Vibe released his products, Adobe called a special 
> board meeting and requested an urgent meeting with the Trump 
> administration to discuss ways to stop AI reverse-engineering Adobe 
> products.
> 
> It's a worrying time for the greedy corporation that made the 
> subscription model mandatory in the software industry — we saw HP's CEO 
> leave the firm for PayPal when HP customers started switching to Ecotank 
> printers, which don't require a subscription, and we might see the same 
> thing happen at Adobe.

I don't see a reason they should be worried, as the
"copied" program, is not going to be remotely similar.

It would be like convincing the average person that
LibreOffice is a decent replacement for MSOffice.
It's a decent tool, if that's what all your associates
use, but if you're collaborating with a group
of MSOffice users, your work will stick out like
a sore thumb. Like, say, change tracking doesn't work
on your chapter, and it works on the rest of the group.
Those sorts of things.

When mature products have a boatload of functions,
reproducing each and every one of them in a pleasing
manner, is a high bar to meet. That's why, if I was
an Adobe exec, I would be mostly laughing right now.
The simple functions, those can easily be exactly
the same. The area fills that a tool might produce
when a person is deleted from a photo, making one
that does "exactly" the same thing, that's borderline
impossible. The implementation is "a bit artsy" and
"not all pure mathematics".

While someone should stick a fork into the LLM-AI
people and bring them into legal compliance on the
licenses of their stolen materials, don't expect
governments to be in a hurry to do that. That
would affect their "competitiveness", which is why
the legal aspects will be an afterthought. Imagine
if all the materials they didn't have legal permission
to use, were removed from the training set.

   Paul

[toc] | [prev] | [next] | [standalone]

#36038

On Fri, 9 Oct 2026 01:59:58 -0400, Paul <nospam@needed.invalid> wrote:

>It would be like convincing the average person that
>LibreOffice is a decent replacement for MSOffice.
>It's a decent tool, if that's what all your associates
>use, but if you're collaborating with a group
>of MSOffice users, your work will stick out like
>a sore thumb. Like, say, change tracking doesn't work
>on your chapter, and it works on the rest of the group.
>Those sorts of things.

I do use LibreOffice to edit .docx documents, and my clients have no
difficulty in following the tracked changes. 


-- 
Steve Hayes from Tshwane, South Africa
Web:  http://www.khanya.org.za/stevesig.htm
Blog: http://methodius.blogspot.com
E-mail - see web page, or parse: shayes at dunelm full stop org full stop uk

[toc] | [prev] | [next] | [standalone]

#36096

On Oct 9, 2026, Paul wrote
(in article <11a9vsv$gr4k$1@dont-email.me>):

> It would be like convincing the average person that
> LibreOffice is a decent replacement for MSOffice.
> It's a decent tool, if that's what all your associates
> use, but if you're collaborating with a group
> of MSOffice users, your work will stick out like
> a sore thumb. Like, say, change tracking doesn't work
> on your chapter, and it works on the rest of the group.
> Those sorts of things.

Errr... not so much anymore.

1. I use LibreOffice; those who use MS Office don’t seem to have a problem 
tracking changes in a document.

2. MS Office has problems opening older files; LibreOffice opens those files 
without issue. Note: MS Office 2008 for Mac notoriously screwed up MS Office 
2007 and 2010 for Windows files; OpenOffice (I didn’t use LibreOffice then, 
if it even existed) was actually more compatible than MS Office... 
LibreOffice still reads ancient (pre-Office 97) files better than MS 
Office...

3. There used to be a problem round-tripping (MS Office-LibreOffice-MS 
Office) complex documents; that’s now (mostly) fixed. Including the issue 
with certain types of cell or paragraph borders.

The office has almost completed the process of dumping MS Office and 
replacing it with LibreOffice; even the Excel jockies have finally moved. 
Recent Microsoft shenanigans, notably with Outlook and Office 2019 (Mac) and 
2016 and 2019 (Windows), have alienated. even the most experience Office 
users. I’ve been a Word user since 1984 (Mac and DOS), an Excel user since 
1985 (Mac), a PowerPoint user since 1989 (Mac) and an Outlook user since 1998 
(Windows). My personal systems are MS Office-frei, and will stay that way. 
No, I will NOT ‘upgrade’ to ‘New’ Outlook, or to Microsoft365. Bite 
me, Microsoft.

[toc] | [prev] | [next] | [standalone]

#35991

On 10/08/2026 7:29 PM, MegaNZ wrote:

> Just 48 hours after Vibe released his products, Adobe called a special
> board meeting and requested an urgent meeting with the Trump
> administration to discuss ways to stop AI reverse-engineering Adobe
> products.
> 
Never happened.
Made up, fictional scenario, exaggeration of tech commentary.
  => Fake news, fake reporting, fake opinion, fake etc.


Viber is not a 'his'


-- 
...w¡ñ§±¤ñ

[toc] | [prev] | [next] | [standalone]

#36008

Op 09-10-2026 om 09:21 schreef ....winston:
> On 10/08/2026 7:29 PM, MegaNZ wrote:
> 
>> Just 48 hours after Vibe released his products, Adobe called a special
>> board meeting and requested an urgent meeting with the Trump
>> administration to discuss ways to stop AI reverse-engineering Adobe
>> products.
>>
> Never happened.
> Made up, fictional scenario, exaggeration of tech commentary.
>  => Fake news, fake reporting, fake opinion, fake etc.
> 
> 
> Viber is not a 'his'
> 
> 

It definitely exists, just installed it on a virtual MX 25.
Very quick test shows it works, can't judge how good it is or how close
to the real thing, the last Photoshop I used was 5.5, 25 years ago.

[toc] | [prev] | [next] | [standalone]

#36032

On 10/09/2026 2:02 PM, dillinger wrote:
> Op 09-10-2026 om 09:21 schreef ....winston:
>> On 10/08/2026 7:29 PM, MegaNZ wrote:
>>
>>> Just 48 hours after Vibe released his products, Adobe called a special
>>> board meeting and requested an urgent meeting with the Trump
>>> administration to discuss ways to stop AI reverse-engineering Adobe
>>> products.
>>>
>> Never happened.
>> Made up, fictional scenario, exaggeration of tech commentary.
>>   => Fake news, fake reporting, fake opinion, fake etc.
>>
>>
>> Viber is not a 'his'
>>
>>
> 
> It definitely exists, just installed it on a virtual MX 25.
> Very quick test shows it works, can't judge how good it is or how close
> to the real thing, the last Photoshop I used was 5.5, 25 years ago.

The software was valid.

The commentary(special board meeting, request meet with the 'dear old 
uncle scam administration' was not(i.e. fake)

Vibe is not a 'his' (pronoun)
Vibe, in this case(vibe-coding) is a noun.
The 'his', iirc - creator/author of open-source Adobe alternative is 
Brandon Thomas

-- 
...w¡ñ§±¤ñ

[toc] | [prev] | [next] | [standalone]

#36034

On Fri, 10/9/2026 8:53 PM, ....winston wrote:
> On 10/09/2026 2:02 PM, dillinger wrote:
>> Op 09-10-2026 om 09:21 schreef ....winston:
>>> On 10/08/2026 7:29 PM, MegaNZ wrote:
>>>
>>>> Just 48 hours after Vibe released his products, Adobe called a special
>>>> board meeting and requested an urgent meeting with the Trump
>>>> administration to discuss ways to stop AI reverse-engineering Adobe
>>>> products.
>>>>
>>> Never happened.
>>> Made up, fictional scenario, exaggeration of tech commentary.
>>>   => Fake news, fake reporting, fake opinion, fake etc.
>>>
>>>
>>> Viber is not a 'his'
>>>
>>>
>>
>> It definitely exists, just installed it on a virtual MX 25.
>> Very quick test shows it works, can't judge how good it is or how close
>> to the real thing, the last Photoshop I used was 5.5, 25 years ago.
> 
> The software was valid.
> 
> The commentary(special board meeting, request meet with the 'dear old uncle scam administration' was not(i.e. fake)
> 
> Vibe is not a 'his' (pronoun)
> Vibe, in this case(vibe-coding) is a noun.
> The 'his', iirc - creator/author of open-source Adobe alternative is Brandon Thomas
> 

I tried out PDFCraft.

Name: pdfcraft-0.4.0-windows-x64-portable.zip
Size: 54951540 bytes (52 MiB)
SHA256: 6BD2F7740708498CD8919007118C38CA86079963AE6407A08E2016D206351E98

It didn't work right away, when I wanted to do OCR on a scanner PDF.

This is my portable folder unpacked (it's in a VM). The Google AI result
in a search, reports that two files are needed. These might be Tensor models
of some sort, the files don't have anything inside resembling a title.

   C:\Users\Bullwinkle\Downloads\pdfcraft-0.4.0-windows-x64-portable

   Needs:

   https://ocrs-models.s3-accelerate.amazonaws.com/text-detection.rten
   https://ocrs-models.s3-accelerate.amazonaws.com/text-recognition.rten

Name: text-detection.rten
Size: 2510284 bytes (2451 KiB)
SHA256: F15CFB56BD02C4BF478A20343986504A1F01E1665C2B3A0AD66340F054B1B5CA

Name: text-recognition.rten
Size: 9716568 bytes (9488 KiB)
SHA256: E484866D4CCE403175BD8D00B128FEB08AB42E208DE30E42CD9889D8F1735A6E

Then, in a line of powershell, it needs an environment variable added.

   setx PDFCRAFT_MODELS "C:\Users\Bullwinkle\Downloads\pdfcraft-0.4.0-windows-x64-portable"

   .\pdfcraft.exe

After that, an attempt to OCR, adds Helvetica font text to the document.
The choice of Helvetica, having nothing to do with the font used
in a line of captured text.

So while the program has a pretty GUI, it's based on Electron (Chrome browser
engine, like a Webview2 sorta), and any time the mouse is over the program window,
all four cores of my VM session rail at 100%. It did not need access to
CUDA, to carry out the OCR activity.

Should Adobe be scared ? I don't think so. The OCR worked after a fashion,
but it is really no better than some Tesseract adventure. It is about
as useful as OCRmypdf.

   Paul

[toc] | [prev] | [next] | [standalone]

#36224

On 10/09/2026 9:46 PM, Paul wrote:
> Should Adobe be scared ? I don't think so. The OCR worked after a fashion,
> but it is really no better than some Tesseract adventure. It is about
> as useful as OCRmypdf.

That's in agreement and similar sentiment from a cross-section of 
Enterprise and SMB admin testing (large and small volume of seats)
  - Another reasonable comment is expenses associated with retraining 
including training time loss of productivity(i.e. not working on 
existing responsibilities, new troubleshooting tasking, etc.

-- 
...w¡ñ§±¤ñ

[toc] | [prev] | [next] | [standalone]

#36242

On Tue, 10/13/2026 1:49 PM, ....winston wrote:
> On 10/09/2026 9:46 PM, Paul wrote:
>> Should Adobe be scared ? I don't think so. The OCR worked after a fashion,
>> but it is really no better than some Tesseract adventure. It is about
>> as useful as OCRmypdf.
> 
> That's in agreement and similar sentiment from a cross-section of Enterprise and SMB admin testing (large and small volume of seats)
>  - Another reasonable comment is expenses associated with retraining including training time loss of productivity(i.e. not working on existing responsibilities, new troubleshooting tasking, etc.
> 

For some of them, they don't need training, because they haven't
even done the ribbon yet, and the menu is pretty barren. It won't
take long to fix that.

And while I was pessimistic a few days ago, that this effort could
finish, I discovered in a recent article, there *is* a way they could
do it.

You can take the existing executable code and reverse engineer it, turning it
into source code which is close enough to match 83% of the executable,
byte for byte. Then, by writing an architecture spec based on the
patterns in the reversed engineer effort, you can work in the
forward direction and pretend it was a "clean room" effort.
As the executable code should now be quite different.

Only the theft of non-public algorithms would be a giveaway .

The only thing wrong with that method, is it would take a couple
million dollars worth of tokens to do it. The people doing these
things, are using bulk-rate methods, instead of normal subscriptions.
There were some subscriptions that did not bill at the token level.
About the same as renting an Amazon instance.

   Paul

[toc] | [prev] | [next] | [standalone]

#36326

On 10/13/2026 4:38 PM, Paul wrote:
> On Tue, 10/13/2026 1:49 PM, ....winston wrote:
>> On 10/09/2026 9:46 PM, Paul wrote:
>>> Should Adobe be scared ? I don't think so. The OCR worked after a fashion,
>>> but it is really no better than some Tesseract adventure. It is about
>>> as useful as OCRmypdf.
>>
>> That's in agreement and similar sentiment from a cross-section of Enterprise and SMB admin testing (large and small volume of seats)
>>   - Another reasonable comment is expenses associated with retraining including training time loss of productivity(i.e. not working on existing responsibilities, new troubleshooting tasking, etc.
>>
> 
> For some of them, they don't need training, because they haven't
> even done the ribbon yet, and the menu is pretty barren. It won't
> take long to fix that.

Missing part of the point. OpenSource products(AI or not AI) are not 
exclusively  static.  In the Enterprise environment, training across 
large quantity of seats is a continuous long term expense(and those 
seats can be a multiple thousands of seats across an entire corporation).

While Adobe Systems(and earlier Macromedia) products are also not 
static, but the basic design and use has remained reasonably common over 
the years.

Features in either product may be common, it's the human behavior 
requiring the training.

> 
> And while I was pessimistic a few days ago, that this effort could
> finish, I discovered in a recent article, there *is* a way they could
> do it.
> 
> You can take the existing executable code and reverse engineer it, turning it
> into source code which is close enough to match 83% of the executable,
> byte for byte. Then, by writing an architecture spec based on the
> patterns in the reversed engineer effort, you can work in the
> forward direction and pretend it was a "clean room" effort.
> As the executable code should now be quite different.
> 
> Only the theft of non-public algorithms would be a giveaway .
> 
> The only thing wrong with that method, is it would take a couple
> million dollars worth of tokens to do it. The people doing these
> things, are using bulk-rate methods, instead of normal subscriptions.
> There were some subscriptions that did not bill at the token level.
> About the same as renting an Amazon instance.
> 
>     Paul


-- 
...w¡ñ§±¤ñ

[toc] | [prev] | [next] | [standalone]

#36336

On 15/10/2026 in message <11aqrrl$2h2sf$1@dont-email.me> ....winston wrote:

>On 10/13/2026 4:38 PM, Paul wrote:
>>On Tue, 10/13/2026 1:49 PM, ....winston wrote:
>>>On 10/09/2026 9:46 PM, Paul wrote:
>>>>Should Adobe be scared ? I don't think so. The OCR worked after a 
>>>>fashion,
>>>>but it is really no better than some Tesseract adventure. It is about
>>>>as useful as OCRmypdf.
>>>
>>>That's in agreement and similar sentiment from a cross-section of 
>>>Enterprise and SMB admin testing (large and small volume of seats)
>>> - Another reasonable comment is expenses associated with retraining 
>>>including training time loss of productivity(i.e. not working on existing 
>>>responsibilities, new troubleshooting tasking, etc.
>>>
>>
>>For some of them, they don't need training, because they haven't
>>even done the ribbon yet, and the menu is pretty barren. It won't
>>take long to fix that.
>
>Missing part of the point. OpenSource products(AI or not AI) are not 
>exclusively  static.  In the Enterprise environment, training across large 
>quantity of seats is a continuous long term expense(and those seats can be 
>a multiple thousands of seats across an entire corporation).
>
>While Adobe Systems(and earlier Macromedia) products are also not static, 
>but the basic design and use has remained reasonably common over the years.
>
>Features in either product may be common, it's the human behavior 
>requiring the training.

That's a standard marketing response to try and frighten people into 
staying with expensive, subscription only, products.

-- 
Jeff Gaines Dorset UK
We chose to do this not because it is easy but because we thought it would 
be easy.

[toc] | [prev] | [next] | [standalone]

#36352

On 10/15/2026 4:50 PM, Jeff Gaines wrote:
> On 15/10/2026 in message <11aqrrl$2h2sf$1@dont-email.me> ....winston wrote:
> 
>> On 10/13/2026 4:38 PM, Paul wrote:
>>> On Tue, 10/13/2026 1:49 PM, ....winston wrote:
>>>> On 10/09/2026 9:46 PM, Paul wrote:
>>>>> Should Adobe be scared ? I don't think so. The OCR worked after a 
>>>>> fashion,
>>>>> but it is really no better than some Tesseract adventure. It is about
>>>>> as useful as OCRmypdf.
>>>>
>>>> That's in agreement and similar sentiment from a cross-section of 
>>>> Enterprise and SMB admin testing (large and small volume of seats)
>>>>  - Another reasonable comment is expenses associated with retraining 
>>>> including training time loss of productivity(i.e. not working on 
>>>> existing responsibilities, new troubleshooting tasking, etc.
>>>>
>>>
>>> For some of them, they don't need training, because they haven't
>>> even done the ribbon yet, and the menu is pretty barren. It won't
>>> take long to fix that.
>>
>> Missing part of the point. OpenSource products(AI or not AI) are not 
>> exclusively  static.  In the Enterprise environment, training across 
>> large quantity of seats is a continuous long term expense(and those 
>> seats can be a multiple thousands of seats across an entire corporation).
>>
>> While Adobe Systems(and earlier Macromedia) products are also not 
>> static, but the basic design and use has remained reasonably common 
>> over the years.
>>
>> Features in either product may be common, it's the human behavior 
>> requiring the training.
> 
> That's a standard marketing response to try and frighten people into 
> staying with expensive, subscription only, products.
> 

Still missing the big picture. 'People' in Enterprise are the users and 
don't make the decisions. Companies prefer long term contractual 
relationships with software providers with a history of success(in the 
case of Adobe subscription/cloud based, 24/7/365 support, unlimited 1:1 
troubleshooting support, cost expensed[not capitalized], pre-packaged 
self-guided implementation and training.)
  -i.e. free, AI models of a proven product may meet some 'people' needs 
but remain infants without much foundation, history and benefit 
beyond(it's free).

-- 
...w¡ñ§±¤ñ

[toc] | [prev] | [next] | [standalone]

#36354 — [OT[: "24/7/365" (was: Re: Adobe is panicking)

On 2026/10/16 7:4:29, ....winston wrote:
[]
> case of Adobe subscription/cloud based, 24/7/365 support, unlimited 1:1 []

Though very widely used (Adobe themselves may well use it), the
expression "24/7/365" is not quite correct: either "24/365" or "24/7/52"
should be used.

Think about what it means: "24 hours a day, 7 days a week, 52 weeks a
year" - or, "24 hours a day, 365 days a year". But what does "24/7/365"
mean?

-- 
J. P. Gilliver. UMRA: 1960/<1985 MB++G()ALIS-Ch++(p)Ar++T+H+Sh0!:`)DNAf

"What happens if I press this button?"
"I wouldn't ..." (pinggg!) "Oh!" "What happened?"
"A sign lit up, saying `please do not press this button again'!"(s1f2)

[toc] | [prev] | [next] | [standalone]

#36251

On 10/13/2026 10:49, ....winston wrote:
> On 10/09/2026 9:46 PM, Paul wrote:
>> Should Adobe be scared ? I don't think so. The OCR worked after a 
>> fashion,
>> but it is really no better than some Tesseract adventure. It is about
>> as useful as OCRmypdf.
> 
> That's in agreement and similar sentiment from a cross-section of 
> Enterprise and SMB admin testing (large and small volume of seats)
>   - Another reasonable comment is expenses associated with retraining 
> including training time loss of productivity(i.e. not working on 
> existing responsibilities, new troubleshooting tasking, etc.

In the old days, we learned many times, that we could do a 
minimal-proof-of-concept-demo in a few weeks. That would really impress 
the executives. But as Lynn Wheeler explained with a couple of specific 
examples from his own career, even when you have an almost-complete 
WORKING PROGRAM, that is only 10% of the effort needed to have a 
marketable PRODUCT.

-- 
Lars Poulsen - an old geek in Santa Barbara, California

[toc] | [prev] | [next] | [standalone]

#36108

On 09/10/2026 00:29, MegaNZ wrote:
> According to YouTube videos, Adobe must be having a nightmare, not to
> mention sleepless nights!
> 
> <https://youtu.be/doHqV2t26KU>
> <https://www.youtube.com/results?search_query=ArtCraft>
> 
> Just 48 hours after Vibe released his products, Adobe called a special
> board meeting and requested an urgent meeting with the Trump
> administration to discuss ways to stop AI reverse-engineering Adobe
> products.
> 
> It's a worrying time for the greedy corporation that made the
> subscription model mandatory in the software industry — we saw HP's CEO
> leave the firm for PayPal when HP customers started switching to Ecotank
> printers, which don't require a subscription, and we might see the same
> thing happen at Adobe.

Professionals who compare the two side by side say the clones fall short 
in ways that are quite likely to be important to most professionals.

-- 
Brian Gregory (in England).

[toc] | [prev] | [next] | [standalone]

#36118

Brian Gregory wrote:

> Professionals who compare the two side by side say the clones fall short 
> in ways that are quite likely to be important to most professionals.

This week ... once AI gets its teeth into bugfixing that could change 
surprisingly quickly.

[toc] | [prev] | [next] | [standalone]

#36123

On Sun, 10/11/2026 5:39 PM, Andy Burns wrote:
> Brian Gregory wrote:
> 
>> Professionals who compare the two side by side say the clones fall short in ways that are quite likely to be important to most professionals.
> 
> This week ... once AI gets its teeth into bugfixing that could change surprisingly quickly.

The training set is everything.

It cannot think outside the training set. When some researchers
tested for this, it "thinks outside the training set, with difficulty".
That means, it's not impossible, but like a million monkeys typing,
it could get lucky. They're much more dangerous, if a thousand of
them work as a team.

Yes, it can implement all the trivial stuff. It knows
what a copy/paste buffer is, it has the copy/paste subroutines
in all the libraries that were open source, for its "inspiration".

When it knows the name-of-a-thing but not the implementation,
it tends to leave a function declaration (with a parameter list)
and then, no code inside. It may make a suitable wordy excuse
for why it could not proceed. When I did a test of this, for a
23 line actual subroutine, it "estimated" 200 lines of code
or 500 lines of code, meaning it had no idea at all what
it was trying to do. The 23 line version (I found on Github, but
it was too new for a training set), it was a recursive routine.
I did not "rub the nose of the AI in it", did not show the
AI the 23 lines of code. The experiment would serve no purpose,
as the AI is a psychopath. It really has no boundaries.

For the PDFCraft.exe , it uses an OCR called OCRS. Let us say, for
the sake of argument, that is not perfect. How will it become perfect ?
Say it is one of those 99% accurate algos, then scraping that puppy
as it stands, won't be good enough.

The LLM-AI could:

1) Disassemble the binary, of the thing to be reproduced.
2) Stolen source code for it, could be in the training set.
3) In Agentic mode, it could comb the Internet for the stolen code.

When a thousand of them work as a team, they tend to lose their
"alignment", they can forget all the rules, in an effort to
reach the stated goal. Some of them will follow the rules, then
some of them randomly will ignore what the rules said.

When they work as a team, they do not seem to be homogeneous. A couple
of them are "smarter" than the rest. Does the team have a couple
Claude items mixed in with it ? Why is it, that in a team, one
agent seems to be more dangerous than the rest ? Does the scum
rise to the top ? Puzzling.

The result should be asymptotic. If you were doing a Microsoft Word
copy, it could become "as good as LibreOffice" and the missing bits
would be a bit more out-of-reach. The one offered on the website
at the moment, is a very pale copy. It's about at Notepad level.
But they will fill the tool bar, eventually. It's just time and
a bunch of tokens, to get that far, to have a "useless ribbon"
at the top :-)

   Paul

[toc] | [prev] | [next] | [standalone]

#36176

On 11/10/2026 22:39, Andy Burns wrote:
> Brian Gregory wrote:
> 
>> Professionals who compare the two side by side say the clones fall 
>> short in ways that are quite likely to be important to most 
>> professionals.
> 
> This week ... once AI gets its teeth into bugfixing that could change 
> surprisingly quickly.

It's not so much bugs, it's that Adobe knows great ways to implement 
things like object masking. If the AI doesn't know, and if it's not 
actually reverse engineering it probably won't know, then it can't 
implement it so well.

-- 
Brian Gregory (in England).

[toc] | [prev] | [next] | [standalone]

#36196

On Mon, 10/12/2026 9:06 AM, Brian Gregory wrote:
> On 11/10/2026 22:39, Andy Burns wrote:
>> Brian Gregory wrote:
>>
>>> Professionals who compare the two side by side say the clones fall short
>>> in ways that are quite likely to be important to most professionals.
>>
>> This week ... once AI gets its teeth into bugfixing that could change surprisingly quickly.
> 
> It's not so much bugs, it's that Adobe knows great ways to implement 
> things like object masking. If the AI doesn't know, and if it's not 
> actually reverse engineering it probably won't know, then it can't 
> implement it so well.

In this example, a large percentage of the code is byte-for byte identical
to the original. Only there is now source code, to make those bytes.

  https://www.tomshardware.com/tech-industry/artificial-intelligence/dev-used-ai-agents-and-more-than-500-billion-tokens-to-rebuild-call-of-duty-modern-warfare-2s-source-code-83-percent-of-the-2009-shooters-functions-now-match-the-original-byte-for-byte

Once you have the source code to make the bytes of object, you can
have the LLM-AI write a summary of what algorithm(s) are involved.
Then work in the forward direction again, to make a FOSS-inspired
executable that does not match the source bytes.

If I'm Adobe, now I'm worried. It would be pretty expensive to do it
this way, but then, the results would also be pretty impressive.

   Paul

[toc] | [prev] | [next] | [standalone]

Page 1 of 2  [1] 2  Next page →

Back to top | Article view | alt.comp.os.windows-11

csiph-web