Posts

Parthenon Random

A common complaint about cryptographically random number sequences is that they do not appear satisfactorily random to human perception. No human can predict what comes next, whether true random or pseudorandom. But it does not satisfy human perception for variety. As the Greeks compensated for the perceived overhang of vertically straight skyscrapers in the architecture of the Parthenon, there exists a way to modify a cryptographically random sequence to satisfy human perception for randomness. In my “Magic Eight Ball” program “Wizard,” published some time ago on www indenturedgeek dot com slash recreations, I implemented the following procedure very satisfactorily: As the stream accepts input from the random stream, we collect a filo array of the last four, five or six most recent responses. Before using the random “suggestion,” we reject it, if it is found in the array. We collect the “n” most recent unique responses, and then add each response we syndicate to the end of the arra...
Post a Comment
Read more

Not all Ai Is LLMs!

The supermajority of Ai applications are uses of LLMs. I published on LinkedIn (no date available) that Ai is not ONLY LLMs. I compared it to Motion Ai. I said LLMs are like language; Motion Ai is like dance. I went further to ask if chemistry Ai (presumably trained on databases of chemical reactions) was like chemistry and if Chess Ai was not likely similar to Chess in some way. One fruitful area of improving a chemistry Ai, or obtaining more data for an initial training model, would be lists of biological metabolites. Where databases are insufficient in Chess, we can compensate by running large samples of Chess Engine “theory” (as used in Chess as a technical term.)  I speculate that patient students could show enterprise by using the PythonSCF module to apply that (Chess) process to chemistry. An early advertisement for Ai was in cpu overclocking, but I speculate that this was actually more likely to have been rule based, like a Chess engine. Obligatory: [also enumerated ...
Post a Comment
Read more

Can LLM’s Benefit From “Proprietary” Training

When composing a prompt for an LLM, it is possible to specify, “in the style of,” both for images and for prose. When training an Ai/LLM can we develop a proprietary quality improvement, by specifying a particular Dictionary (some specific printing of Webster’s or the OED) to include the carefulness of defining language and the creativity and variety [aka “diversity”] of the example usage sentences ? For another application of the same idea, what is the result of including Cicero, Caesar, Montesquieu, Machiavelli and Tocqueville in specific translations and printings? Expanding on this concept, Jonathan Swift, Daniel Defoe, R. L. Stevenson, Earnest Hemingway and Mark Twain would each offer a unique characteristic. Do we improve or adulterate, when we “build on that success,” by including more gross content, or cross sections for variety?
Post a Comment
Read more

Do Dictionary Specifications Give Us Insight Into Training Neural Nets?

When we purchase a Dictionary, we can specify “complete,” v “exhaustive.” Exhaustive specifies that every word in every definition is redundantly defined within the same catalog. 100 % Redundant is another metric. Does this give use insight into training Neural Nets?
Post a Comment
Read more

Can Directed Search Be Trained To Compare And Contrast?

I asserted that agnostic layer may improve directed search. I might have been more correct to say that Brin and Page brought us “search,” rather than to say Brin and Page  brought us “directed search.” The idea of comparing and contrasting opposing ideas rather than enhancing specialization seems to rest on the axiomatic concept of “Not.” It might difficult to assign a definition or an idea a truth value. However, an LLM suggests a dictionary, and a dictionary suggests antonyms as well as synonyms. For this reason, I want to suggest using a thesaurus or a dictionary to give weighted associations to antonyms.
Post a Comment
Read more

Is Paradox NP-Complete?

 The notorious reputation of NP Completeness promises to resolve decidabilities for specific cases.  With the word/expression “NP IN-complete,” as a clue, we can ask if it is worth considering a connection to the more famous incompleteness observation, Goedel’s Incompleteness Theorem. My preferred enumeration is “No finite set can contain all its own axioms.” To mitigate potential ridicule and cover my anatomy, this is a student type research blog, not a publication.
Post a Comment
Read more

Agnostic Layer

When Brin and Page brought us “directed search,” I fought for “agnostic search,” as a vocabulary term and a choice. I found Google to be receptive and responsive as a corporation, and Chloe Condon brought us the concept of agnostic layer. I wholeheartedly believe that agnostic layer will improve directed search. With any luck, it can also introduce opposing views alongside agreeing views that are more specialized, or enthusiastic. This article is for attribution, and to thank Chloe Condon for agnostic layer. I’m holding out for DuckDuckGo, Dogpile, and agnostic search as a choice.
Post a Comment
Read more