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

Groups > sci.physics > #895367

Visual Perception helps reading Matrices, Fractions, etc.. (Re: Amazing Visual Perception, even for ASCII Input [GPT 5.3])

Cross-posted to 2 groups.

Show all headers | View raw

Hi,

Maybe some 2D Diagrammtical Reasoning concerning
ASCII Texts , is result of training the AI on
corpus of math texts that contains usually a

lot of 2D typography. One can test the following.
I was juggling with Smarandache Dual numbers, so
I asked ChatGPT and DeepSeek about the following epsilon:

(1 -1)
(1 -1)

------- ChatGPT Protocoll begin ----------------

Q: Isn't this a Smarandache Dual number epsilon:

(1 -1)
(1 -1)

A: /* It computes A^2 which gives 0 and concludes: */
So your matrix is a valid matrix model of the
dual unit, just not the canonical one.

------- ChatGPT Protocoll end ----------------

------- DeepSeek Protocoll begin ----------------

Q: Isn't this a Smarandache Dual number epsilon:

(1 -1)
(1 -1)

A: /* It computes M^2 which gives 0 and concludes: */
Could represent ϵ itself if one redefines the basis
for dual numbers in the algebra of 2×2 matrices
generated by M and I, but that would not match
standard textbooks’ notation.

------- DeepSeek Protocoll end ----------------

But ChatGPT and DeepSeek are not omnicient enough
to even find the source of A resp. M:

https://fs.unm.edu/DualNumbers.pdf

So they didn't do retrieval. They did some little
mathematical reasoning on their own.

Have Fun!

Bye

Mild Shock schrieb:
> Hi,
> 
> The amazing things is ChatGPT, and possibly
> other LLMs don't think linearly. I literally
> posted the code like this:
> 
> 040C AF
> 040D 11 0000
> 0410 CD3300
> 0413 21 0000
> 0416 36 00
> 0418 E5
> 0419 D1
> 041A 13
> 041B 01 0904
> 041E ED B0
> 0420 3E 02
> 0422 32 0000
> 
> XOR A
> LD DE,0000h
> CALL 0033h
> LD HL,0000h
> LD (HL),00h
> PUSH HL
> POP DE
> INC DE
> LD BC,0409h
> LDIR
> LD A,02h
> LD (0000h),A
> 
> Two consecutive columns. And it moved
> the columns side by side all on its own.
> I noticed that many times,
> 
> that modern transformer AI architectur
> has quite amazing visual perception,
> even for ASCII input.
> 
> Have to double check whether it already
> used GPT 5.3, or only GPT 5.2:
> 
> Introducing GPT‑5.3‑Codex
> Expanding Codex across the full spectrum
> of professional work on a computer
> https://openai.com/index/introducing-gpt-5-3-codex/
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> Even ChatGPT can recognize it:
>>
>> 040C AF        XOR A
>> 040D 11 0000   LD DE,0000h
>> 0410 CD3300    CALL 0033h
>> 0413 21 0000   LD HL,0000h
>> 0416 36 00     LD (HL),00h
>> 0418 E5        PUSH HL
>> 0419 D1        POP DE
>> 041A 13        INC DE
>> 041B 01 0904   LD BC,0409h
>> 041E ED B0     LDIR
>> 0420 3E 02     LD A,02h
>> 0422 32 0000   LD (0000h),A
>>
>> It responds with:
>>
>> This is assembly language / machine code for a Zilog Z80
>> CPU (or a very close Z80-compatible processor). 🧠
>>
>> The clues are the hex opcodes and instructions like:
>>
>> AF → XOR A
>> CD → CALL
>> 21 → LD HL,nn
>> 32 → LD (nn),A
>> ED B0 → LDIR
>> CB 1C → RR H
>> E5 / D1 → PUSH HL / POP DE
>>
>> These are classic Z80 instruction encodings.
>>
>> Bye
>>
>> https://chatgpt.com/share/69ad3f1e-0c68-8013-8cc3-b3f43267f69e
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> Ross Finlayson schrieb:
>>>  > Three decades in software engineering helps read code.
>>>
>>> Thats not much. Given that I wrote
>>> an Euler Number computation to 1000
>>> digits in Z-80 Assembler when I was
>>>
>>> < 13 years old, I have > 5 decades
>>> of software engineering.
>>>
>>> LoL
>>>
>>> Bye
> 

Back to sci.physics | Previous | Next — Previous in thread | Next in thread | Find similar

Thread

Re: AI solves probl;em Knuth was/is working on! Mild Shock <janburse@fastmail.fm> - 2026-03-04 08:29 +0100
  Terrence Tao: On Jan 4, ChatGPT was able to produce a proof Erdos #728 (Re: AI solves probl;em Knuth was/is working on!) Mild Shock <janburse@fastmail.fm> - 2026-03-07 11:58 +0100
    Aristotle: The Era of Vibe Proving is Here (Re: Terrence Tao: On Jan 4, ChatGPT was able to produce a proof Erdos #728) Mild Shock <janburse@fastmail.fm> - 2026-03-07 12:08 +0100
      Re: Aristotle: The Era of Vibe Proving is Here (Re: Terrence Tao: On Jan 4, ChatGPT was able to produce a proof Erdos #728) Ross Finlayson <ross.a.finlayson@gmail.com> - 2026-03-07 07:21 -0800
        Re: Aristotle: The Era of Vibe Proving is Here (Re: Terrence Tao: On Jan 4, ChatGPT was able to produce a proof Erdos #728) Ross Finlayson <ross.a.finlayson@gmail.com> - 2026-03-07 07:36 -0800
          Only, three decades in software engineering? Re: Aristotle: The Era of Vibe Proving is Here (Re: Terrence Tao: On Jan 4, ChatGPT was able to produce a proof Erdos #728) Mild Shock <janburse@fastmail.fm> - 2026-03-08 10:00 +0100
            Re: Only, three decades in software engineering? Re: Aristotle: The Era of Vibe Proving is Here (Re: Terrence Tao: On Jan 4, ChatGPT was able to produce a proof Erdos #728) Mild Shock <janburse@fastmail.fm> - 2026-03-08 10:19 +0100
              Amazing Visual Perception, even for ASCII Input [GPT 5.3] (Was: Only, three decades in software engineering?) Mild Shock <janburse@fastmail.fm> - 2026-03-08 10:34 +0100
                Visual Perception helps reading Matrices, Fractions, etc.. (Re: Amazing Visual Perception, even for ASCII Input [GPT 5.3]) Mild Shock <janburse@fastmail.fm> - 2026-03-08 20:21 +0100
      Re: Aristotle: The Era of Vibe Proving is Here (Re: Terrence Tao: On Jan 4, ChatGPT was able to produce a proof Erdos #728) x <x@x.net> - 2026-03-08 19:10 -0700
        Biobrain cannot digest "Artificial" in AI (Was: Aristotle: The Era of Vibe Proving is Here) Mild Shock <janburse@fastmail.fm> - 2026-03-09 11:21 +0100
          Prolog vendors don't understand Tabling anymore (Re: When Supercomputing discovered BFS) Mild Shock <janburse@fastmail.fm> - 2026-04-04 14:10 +0200
            Buddos Waterloo: Checkpoint McCune (Re: Prolog vendors don't understand Tabling anymore (Re: When Supercomputing discovered BFS) Mild Shock <janburse@fastmail.fm> - 2026-05-10 16:20 +0200
              Could ISAMORE anti-unification help? (Re: Buddos Waterloo: Checkpoint McCune) Mild Shock <janburse@fastmail.fm> - 2026-05-10 16:48 +0200

csiph-web