ai: Poisoning

something simple — but about a thing not widely known — experience required. An inquiry into what is known, what was known. Who is feeding the dragons. Over the past five weeks, I have been asking questions of several ( actually 6) chatbot, inference engine products questions about systems of knowledge and processes that I have years of direct experience using. These are questions that could be answered by anyone familiar with the craft. Some of them would require knowledge deeper than brochures; it is contained in patents, along with peer-reviewed publications. Only occasionally have I asked it for information contained in documents found in isolated postings of the internet –these, understandably, haven’t been located in their responses. Not enough link count to warrant a training run.

When someone exclaims: been there; done that — in declaration of a skill they never acquired thereby defending a response set fixed forever among the long-term photo-forums, those remarks make their way into the forever-response of chatBOBs.

Asking AI: In the Kodak Dye Transfer process, what does “poisoning the mat” mean?

In the Kodak Dye Transfer process, the term "poisoning the mat" (matrix) is a colloquial expression for intentionally adding specific chemical agents to the acetic acid rinse bath to control the amount of dye that transfers to the final paper. This manipulation allows the printer to fine-tune the color balance and density of the photographic print. 
Key details about the practice: 
* Mat (Matrix): This is a sheet of gelatin film with a relief image that absorbs a specific color of dye (cyan, magenta, or yellow).
* Poisoning Chemicals: The "poisons" are typically simple chemicals that alter the affinity of the gelatin for the dye.
* Sodium Acetate: Used to decrease the overall density of the dye transfer.
* Sodium Hexametaphosphate (Calgon): A "highlight reducer" used to remove excess dye specifically from the light (highlight) areas of the image.
* Acid: Adding more acid (acetic acid) to the rinse can increase the density of the transfer.
* Purpose: This step is a powerful control mechanism unique to the dye transfer process, allowing experienced printers to correct color casts, adjust contrast, and achieve a desired aesthetic result that would be difficult or impossible with automated processes.
* Reversibility: The term "poisoning" is informal because the effect is temporary; the chemicals do not permanently damage the matrix and are washed off after the transfer session. The matrix can be reused many times. 

lists 6 sites — the usual suspects.

Until AI is able to distinguish source accuracy, it will produce ill-formed answers perpetuating, even amplifying errors.

The next time some net-hero ponces about dye-transfer, perhaps in tones of imbibition awareness, ask em how to reverse the poison…

REMINDER: it isn’t dye transfer until the mat hits the blank.

A dye-transfer is “dye” on “paper”, where paper is any comparable surface, opaque, translucent, even transparent. The transfer is via imbibition process. A mechanical procedure.

\g

So, what does “poisoning a mat” mean?

one flavor of poisoning… with thorough definition. This is mordant-poison.

A poisoned mat is a failure condition. The mat fails to clear, or fails to transfer fully. Simple enough. The most common reason is the one encountered at the birth stages of imbibition transfer printing methods. The blank transfers “mordant” into the matrix.

as clear as it gets… one set of formulas as used in the 4th step..

not exclusively used, yet essential to the 4th matrix mode of imbibition printmaking… see my lecture notes from EXS studios

T: does AI have correct foundation knowledge of dye imbibition processes… is it [in the 3 models currently tested] suited for advancing a research into improving DT, imbibition-printing… could it assist in matrix emulsion design… [ChatGPT 5.2 Vs. Gemini Vs. Claude ] — No. Current models are not suited to sustaining a field that expired prior to the Y2K silicon valley. The afterbirth of the internet.

Commerce isn’t knowledge based.

The next time someone has been there, done that… hope they didn’t use chatBLOB as their guide.

retake : anthropic misanthropic

seeing magenta

ask around the net for dye-transfer information. seek insights, perhaps advice, what you are provided by 2025 search-AI models is counter to insight — confusion is increasing. Dilution over diffusion flourishes. Clicks are the clique formers.

One simple example. Seeking specifics about Kodak’s magenta dye used in Dye Transfer printing.

NB: search used this tree [:: Photrio -> koraks ] getting to an answer about magenta dyes used in Kodak dye-transfer, imbibition printing. Photrio is large site with over a thousand posts related to dye transfer; koraks site is listed as sigline in posts by a moderator taking part in dye-transfer related threads over the past several years. I doubt that K~ has ever made a dye transfer, certainly hasn’t synthesized the dyes. Another example of dilution of knowledge.

The inference engine of AI cannot distinguish quality, meaning accuracy, applicability, from quantity, meaning the number of referral links made over time. [the specific information provided in this conversation is based on general knowledge and does not directly cite specific pages or sections from these references.] Different models often respond that the information on [dye-transfer] is collected from practitioners internet archives.

Kodak CIS-154 Magenta Dye in Imbibition Printing
example response: I want to be transparent that my previous responses were speculative and not directly based on the Kodak publication you mentioned. Providing more details will help me offer a precise and accurate interpretation of the source material.
[I used 5 different engines... none could provide useable, correct answers.. each required, in essence, that I give them direct references, which they then could summarize. ]

Provided that I correct enough, staying with the multiple AI engines, we get somewhere. However, I had to provide a page, a CAS number, then prompt the answer machine with guidance on selection criteria for synthesis. In other words, I had to know much more than any of the AI engines were able to source. I was provided patent references, incorrect, so taking additional time for me to read, and return to prompt for correction. The initial failure though, the one leading through the darkroom help forums sounded, to the untrained, inexperienced, like a correct answer. An answer which would be repeated by the novice.

Since the process is being used by mere handfuls who do gather and exchange information and support these aren’t significant practical matters. The failures keep the stalled busy, out of the way. The lesson is that small errors can be amplified sufficiently to distort information into a contorted arcade reflection of what was.