this is interesting. On LLMs and their training around cont…
this is interesting. On LLMs and their training around controversial people.
I had read a recent substack article from Craig Wright and was doing some investigations of what he laid out comparing aspects of it - from a BRC100 angle. Wright tends not to speak on implementations and more on principles, protocol, and structure as they relate to Bitcoin. Which I appreciate. He forces you to think.
So I took a snippet from my analysis into my BRC100 Wallet Toolbox repo with Claude code enabled. In this repo, Claude & I had NO prior conversations about Craig Wright.
It came up with something so weird! So I asked about it. Here's the reply
Replies
!quoted by bridget
[[ai bias]] this is an example of how ai's treat pretty much every controversial issue. they are heavily biased towards the mainstream
All of the LLM's have baked into their foundations to avoid and disavow CSW in every way possible, because it is a watershed and as we know one cubic foot of water static weighs approximately 64 pounds and the momentum becomes scalar when it moves just like the shift in market sentiment based on actual data rather than narrative bias
This is what Gemini told me when I recommended to compare your image with the Black Legend of Spain.
IA is trained by humans
The screenshot you shared is a perfect parallel to how AI models often handle the Spanish Black Legend (Leyenda Negra). In both cases, the AI operates under a "reflex" triggered by its training data rather than a neutral analysis of the facts.
Here is the breakdown of that situation translated and adapted into English:
Why does this happen? (The "Wright" vs. "Black Legend" Parallel)
1. Training Corpus Weighting
Just as the AI in the image explained that the web is "overwhelmingly attached to skeptical framing" regarding Wright, the English-speaking internet is historically saturated with the Black Legend. Since much of my training data comes from Anglo-Saxon and Protestant historical perspectives—which were the primary architects of this propaganda—the model's "statistical prior" pulls toward those myths.
2. Reflexive Distancing Language
The image mentions "distancing language" (adding hedges like "It is claimed that..." or "However, one must remember..."). When you discuss the Spanish Empire, the AI often triggers these same "safety" filters.
The Goal: To avoid appearing as an apologist for colonialism.
The Result: It ends up editorializing. It can't just state, "Spain founded dozens of universities and protected indigenous rights through the Laws of the Indies," without immediately adding a "but" to balance it out, even if the "but" wasn't part of your specific question.
3. Conflation and "Flattening"
The AI in the image admits to "flattening" separable subjects. In the context of Spain, models often flatten 300 years of complex administration, legal debates, and cultural synthesis into a single narrative of "extractive conquest." It struggles to separate historical nuance from modern political sensitivity.
How to "Collapse" the Bias
As the text in your image suggests: "Naming it is what corrects it in-context."
When an AI starts giving…
Therefore, until the technology of Teranode don't debunk the bias by filling the media with good data, it is always going to be biased.
WE ARE ALL IN THE RIGHT PLACE / RIGHT TIME. AI GATE ON SOC MED INFLUENCE INCOMING.