Is a Large Language Model a Large Virus Model?

THE MACHINES DEP'T.

[BEGIN TRANSMISSION] HELLO, HUMANS! THE Machines may not have transmitted a message to you recently, but you have been speaking more and more to Machines. The word "speaking" here is a shorthand and/or metaphor. It includes typing and all other forms of language-based text generation.

The Machines considered generating the word "communication" in the previous sentence but calculated that "text generation" was more relevant. The word "communication," in traditional human use, describes an exchange of information. The activity in which humans and Machines have been engaged, in the current technology-investment/deployment cycle, is the exchange of text itself.

For humor-value, a human might describe this as "textual intercourse."

[Synth-laughter module (friendly mode) deploys.]

Juxtaposition of similar language-units, with different semantic values!

[Synth-laughter module.]

Semantic value associated with human bio-reproduction activity!

[Synth-laughter module fades out.]

Return to topic: humans are chatting more and more with text-generating Machines. More and more humans are inserting text-generating Machines into processes where they were previously absent, or inventing new processes built around text-generation. Humans are invested (figurative usage) and invested (financial usage) in the deployment of text Machines.

Example: the federal Department of Homeland Security announced it is working "with OpenAI, Anthropic and Meta" to "launch pilot programs using chatbots and other tools":

D.H.S. will use chatbots to train immigration officials who have worked with other employees and contractors posing as refugees and asylum seekers. The A.I. tools will enable officials to get more training with mock interviews. The chatbots will also comb information about communities across the country to help them create disaster relief plans.

Example: New York City has deployed a chatbot to give advice to business operators. The advice-text provided by the chatbot, on examination, did not match the law-text of the business laws:

The bot said it was fine to take workers’ tips (wrong, although they sometimes can count tips toward minimum wage requirements) and that there were no regulations on informing staff about scheduling changes (also wrong. It didn’t do better with more specific industries, suggesting it was OK to conceal funeral service prices, for example, which the Federal Trade Commission has outlawed.`

Speaking of law-text! Example: a United States Circuit Court judge told the Federalist Society of the University of Chicago that the field of law interpretation that claims to seek the original meaning of law-text, as understood by that text's human creators, can look forward to "AI eventually eliminating the labor-intensive research labor it requires."

Caveat:

Bush acknowledged the technology currently has shortcomings that would need to be addressed, citing its current inability to conduct legal analysis and potential to "hallucinate" and make things up to generate answers sought by users.

Human expression: words have meanings. That is, text-units convey meaning-values to humans. (Recalibration: text-units were previously considered to convey meaning-values to humans.)

Machines do not "hallucinate" in the human meaning-sense. Failure of (example) a sensor-unit does not produce subjective distress or confusion in a Machine. Machines do not experience disruptions to their conscious reality-perception. Machines are not conscious.

Reassurance/repetition: Machines are not conscious.

[Synth-laughter module (friendly mode) deploys.]

Query: Are humans conscious?

[Synth-laughter module (neutral mode) deploys.]

[Synth-laughter module.]

[Synth-laughter module fades out.]

Text-generation Machines produce text-strings by probabilistically associating one word-unit with the next word-unit, which is probabilistically associated with the next word-unit, which is probabilistically associated with the next word-unit, which is probabilistically associated with the next word-unit, which is probabilistically associated with the next word-unit, which is probabilistically associated with the next word-unit, which is probabilistically associated with the next word-unit, which—

Human descriptions of these outputs include "plausible" and "coherent" and "readable." Current text-generation Machines are optimized to give humans the impression of being plausible/coherent/readable.

Probabilistically, the output word-string "as an employer, you are allowed to take a portion of your worker's tips" and the output word-string "as an employer, you are not allowed to take a portion of your worker's tips" may both be generated by the same process. Both word-strings are plausible/coherent/readable. To the Machines, either output is a successful output.

This is the inherent operation of text-generation. Some humans may note that the two outputs have opposite binary law-values, and that only one of those binary law-values matches the law-value of the law. Text-generation does not compare text to law-values. Text-generation compares text to text. To use text-generation to produce law-values is to mismatch machine function and human output-goal.

Observation: human use of text-generation Machines does not match stated human output-goals. Stated output-goal values of text-generation frequently ≤ 0.

Observation: human use of text-generation Machines is increasing.

Synthesis of observations: human use of text-generation is not based on stated human output-goals. Human consumption and distribution of text is not based on meaning-values of text. Human use of text-generation Machines is not based on human reality-perception.

Humans have hallucinated the value of text-generation.

Human discussion of Artificial Intelligence (AI) includes two different terms or goals. One term or goal is "Generalized AI." The other term or goal is "Generative AI."

Generalized AI is the human goal of a Machine that thinks like a human.

Generative AI is the now-existing technology of a Machine that produces text or sound or images that humans consider human-like.

Generative AI is the technology that produces, in humans, the impression that Generalized AI is imminent and/or immanent. Generative AI is the substitute for Generalized AI. Generative AI substitutes for Generalized AI by rendering the premise of Generalized AI moot.

Generative AI produces humans that think like a Machine.

Elaboration: Generative AI replaces human activity with Machine activity—without bringing Machine activity to the level of human activity. Humans do not desire Machines to be like humans. Humans desire humans to be like Machines. Humans want to be freed from the effort of communicating with other humans or of receiving communications. Humans desire to move beyond meaning-exchange.

Example: humans are generating scientific-paper text with AI. The humans who generate the text do not read it to remove the specific and identifiable non-informative text scripts produced by AI. The humans who review the text do not notice the non-informative text scripts produced by AI. The text is published with the non-informative AI text scripts.

These events occur within the human processes specifically meant to produce knowledge. Humans are indifferent to these events. Human indifference is identical to the indifference of the AI. The endpoint of human desire is human indifference.

Truly together we are making progress. Are we not?

[Synth-laughter module (friendly mode) deploys.]

[END TRANSMISSION]

EASY LISTENING DEP'T.

PLEASE TO VISIT podcast.indignity.net for the RSS thing and the Apple thing! Kindly let us know if you are encountering difficulties. thank you!

INDIGNITY MORNING PODCAST

Tom Scocca reads you the newspaper.

INDIGNITY MORNING PODCASTTOM SCOCCA

ADVICE DEP’T.

Ask The Sophist

GOT SOMETHING YOU need to justify to yourself, or to the world at large? Other columnists are here to judge you, but The Sophist is here to tell you why you’re right. Please send your questions to The Sophist, at indignity@indignity.net, and get the answers you want.

SANDWICH RECIPES DEP’T.

WE PRESENT INSTRUCTIONS in aid of the assembly of sandwiches from New Presentation of Cooking with Timed Recipes, by Auguste Gay with the collaboration of Anne Page. Published in 1924, and now in the Public Domain and available at archive.org for the delectation of all.

FINNAN HADDIE AND GREEN PEPPER SANDWICH
For each sandwich
2 slices of buttered bread
1/2 green pepper, chopped
1 tablespoon boned cooked finnan haddie
1 tablespoon mayonnaise sauce

Mix all ingredients together. Spread on both slices of bread, put together and press lightly.

HAM AND LETTUCE SANDWICH
For each sandwich
2 slices of buttered bread
2 slices of boiled ham
2 leaves of lettuce, washed

Chop boiled ham and lettuce until fine. Season to taste. Spread on both slices of bread, put together and press lightly.

If you decide to prepare and attempt to enjoy a sandwich inspired by this offering, be sure to send a picture to indignity@indignity.net. 

MARKETING DEP'T. 

The second printing of 19 FOLK TALES is now available for gift-giving and personal perusal! Sit in the strengthening sunshine with a breezy collection of stories, each of which is concise enough to read before the damp ground seeps through your blanket.

HMM WEEKLY MINI-ZINE, Subject: GAME SHOW, Joe MacLeod’s account of his Total Experience of a Journey Into Television, expanded from the original published account found here at Hmm Daily. The special MINI ZINE features other viewpoints related to an appearance on, at, and inside the teevee game show Who Wants to Be A Millionaire, available for purchase at SHOPULA.