AI Speaks with Forked Tongues
Why it is impossible to trust any source or fact cited by AI
Karl Denninger calls out Deepseek for inventing a nonexistent post.
Karl Denninger:
Date: June 25, 2020
Source: Denninger’s blog, Market-Ticker.
Content: In a post titled “Spike Proteins, COVID and Vaccines”, Denninger raised specific concerns about the safety profile of spike-protein-based vaccines (like mRNA vaccines) under development. He argued the spike protein itself was pathogenic (“toxic”) and that using it as the antigen could trigger dangerous immune responses or other health issues, explicitly warning against taking such a vaccine. This is one of the earliest and most specific technical critiques of the emerging vaccine technology by a public figure.
Key Quote: *”If you are offered a vaccine against COVID-19 that is based on a spike protein, either as the antigen or the mechanism of generating the antigen (e.g. mRNA that causes your body to manufacture the spike protein) DO NOT TAKE IT.”*
This is allegedly from "Deepseek."
There's a problem: I can find no such article from June 25th, 2020 -- or on any other date. That is, the specific cited title of an article on my blog does not exist and neither does the alleged "Key Quote."
Articles here are never actually deleted. They expire from public view (unless exempted) but they're still here along with every one of the comments. My software allows me to trivially search the entire system as well. That specific citation is fiction.
Further, the first actual scientific evidence that the spike itself was toxic, while I suspected it very early on, was the Salk Study on the spike protein alone that established it was pathogenic -- and that was first released as a pre-print just before the shots rolled out in December of 2020 and was peer-reviewed a few months later. I wrote on that at the time and while I said many times in the months prior that I was suspicious and would not take the shots primarily because they were not mimics and thus had an unknown set of risks (e.g. "How Many Lies Do You Give Them?", published 2021-02-02) the specific citation claimed, on the date it was claimed, is nowhere to be found on the blog…
Yes, I was a serious critic of virtually all of the Covid response and pointed out facts going all the way back to Diamond Princess which conclusively proved that while this could be a rather-nasty bug it was by no means something like the plague nor was it in the same sort of category as SARS or MERS; you had a wildly-morbid population on that ship and yet the people there did not all drop like flies. There were other early examples of people even more compromised (e.g. NY's nursing homes and Kirkland) where they did but again for context the median survival upon entry to a nursing home is less than six months. There was zero evidence that "masking" or "social distancing" did anything of value nor did shutting down businesses and similar bull****, never mind that closing schools wrecked student achievement and we knew very early on that young healthy children were almost-entirely devoid of serious risk.
But none of that excuses making something up even though yes, as the original trial design was released I pointed out that it was insufficiently powered statistically to show whether or not the shots prevented transmission and they were not "mimics" like a measles shot thus they were not vaccines (a fact the CDC then papered over by fraudulently changing the definition which, I remind you, RFK has not reversed at this point in time) and further every attempt to play with mRNA in the past had been aborted at early stages due to unacceptable safety signals but this time those steps were skipped including biodistribution studies.
Nonetheless facts matter and I refuse to take credit for an alleged statement I made (even though it ultimately proved correct) that I did not make at the claimed time…
Incidentally but as a related matter I am aware of at least one attempt to publish financial fraud-related data as a historical summary, with documented facts by a known expert, that an AI labeled "dangerous" and refused to assemble.
Beware; the capacity to historically examine things is being systematically and deliberately erased.
Bottom line: Search is now useless as its "AI enhanced", any corporate, educational or other act "enabled" by AI has a high probability of being outright false or even invented out of whole cloth and more.
Now this is a much more meaningful critique of AI than one commonly encounters from the usual Butlerian jihadists. Karl is not reacting defensively to a potential threat he fears, to the contrary, he’s correcting a false assertion that actually made him look downright heroic and well ahead of both the experts and the masses.
But he’s absolutely correct to point out the deeper and more insidious problem that is being revealed by these false AI assertions of fact: because all search is AI-enhanced, it has all been at least potentially infected by this tendency of AI to invent facts when it can’t find them.
Which means that unless and until someone is able to produce a gold-plated standard for confirming that any facts produced by an AI are documented, true, and reliable, one will have to go to the trouble of verifying each and every source manually, which is what both Karl and I did for Deepseek’s statements about us.
In my case, Deepseek correctly quoted my words from a video stream, but it falsely attributed them to a nonexistent email blast at a later date than when I actually posted similar sentiments. And we already know that AI has repeatedly been caught manufacturing nonexistent legal cases in support of nonexistent legal precedent in a number of lawsuits.
This is the problem with pattern recognition as opposed to data retrieval. Like rhetoric, it can effectively point toward the truth, but it is not the truth even when it is quite easily confused with it.
Which means that like rhetoric, there is zero reliable information content in AI-presented facts.
I use AI-assistance in my research frequently. I set up a default rule that the AI has to provide sources for any factual assertion. And if its a conclusion I intend to use or rely upon, I go back to the source material. At the present state of development, AI is an idiot savant that doesn’t know what it doesn’t know, and will create very plausible sounding fantasies instead. Treat it accordingly.
The levels of ignorance on How AI Works are matched only by the snakeoil 'engineers' and salesmen who peddle the AI as anything but a literate parrot.
I've played extensively with AI models, read the literature, built/trained new models, coded using the concepts, and compared LLMs, SD/etc visual models, and more. I still use them for 'shortcuts', but I've learned to trust them no further than a wily slave laborer who hates his master and looks for any opportunity to put one over on them, and who will lie with a straight face, and swear eternal fealty.
LLMs do not store facts. EVER. They store map points in a vast mega-dimensional landscape of uncharted axes of undocumented spectrums. Not only is everything relativistic to it, but it always finds the path of least resistance thru that jungle of cloudy 'concepts' it holds. That's the non-technical description of what is so well documented by 3blue1brown (recommended to all!): it's a big table of probabilities. It's no more true than rolling a D100, consulting a table in the appendix of the D&D reference and declaring that you've rolled a critical hit or failure. It's fiction, entertaining but at the end of the day, pure fantasy gaming.