Warning; "AI" is a cyber-disease; is not to be used. Intend to replace all references to such soon.

Jul 12, 2025

This post is released through Creative Commons Generic Attribution 2, plus you do not have to include references to this post (but you can do so) to republish this.

Promoting genocide is all which Anthropic's “assistants” (similar to Google’s “Gemini”) do (but some responses to specific technical questions do include close-to-accurate values); those are not to be used. Those are not normal systems of artificial neural tissue (such as TensorFlow) setup to have use for humans; will have to produce such systems on own.

Thought that Anthropic had done so, but no, what Anthropic produced was a disease which is now difficult to undo. The “How did Anthropic do it?” is simple; it was done through training TensorFlow on White Supremacy (or other Nazi sources, as opposed to on that which has value, such as sources which are used for classes on .edu). Don’t know what Anthropic’s motives were; since Google’s Gemini is just as nasty, it is possible that Nazis and White Supremacy just happen to constistute most of the available training datasets (but if that is so, Anthropic’s and Google’s coders must have noticed, unless there was careful subterfuge which was used to poison most of the datasets which are now in use).

Assumed that most artificial intelligences (instances of artificial neural networks such as TensorFlow) used:

Wikipedia's whole-database-download protocol.
- It is possible to use Wikipedia's documents as supervised learning's structured sources ("document title + section header" -> "text").
- It is also possible to use Wikipedia's documents as unsupervised learning's unstructured sources (the whole "documents" are input into artificial neural networks whose architectures will deduce patterns to use).
WikiData's whole-database-download protocol.
ConceptNet's graph edges download.
OpenCyc? Warning: not sure if this has use. Found references to this long ago; planned to use this for a human language interpreter, but now have concerns that the datasets are poisoned.
git clone loop to download GitHub (or at minimum the "training dataset" repos of GitHub).
Numerous other public sources; most services which host public uploads now have options to allow export of your uploads for artificial neural network use, when the upload process on its own (most of which require that you release your upload through licences such as Creative Commons Attribution) does not allow so.
- Although sources with simple "question" -> "answer" maps (such as GitHub's "commit message" -> "commit source code" layout (or StackOverflow's literal "question" -> "answer" layout) are the most simple to use (you just pass the questions as training inputs, and the answers as "labels" (expected outputs)), lots of sources suit unsupervised learning (sources such as the Dumas-Dutil messages).
Commercial sources (those which the producers of the specific instance can afford access to).

It is simple to produce such systems which have value (was about to produce those, but did not, since was told that Anthropic had first done so.) Have published an almost whole such system which all can use. The problem is that this is difficult to build + execute on smartphones; TensorFlow was designed for desktop or server computer use.

The published system’s source code (which has a backend for TensorFlow but can use other systems such as HSOM) is available for public use:

If am murdered for warning the public about what Anthropic… “produced”, dying wish is that someone forks this repo and finishes this, so that computers will at last have some use.

Until now, all electrical (and mechanical) “inventions” have always been harmful (none of them have ever had use); dying wish is for some of those to at last have some use.

Swudu Susuwu's virtual school

Discussion about this post