In case you do not keep up with the various shenanigans associated with Abominable Intelligence, an absolutely hilarious story broke a few days ago when X (the new and improved Twatter, after its acquisition by Brolon MuZk a few years back) updated its in-house AI model, Grok. The update apparently resulted in a large language model (LLM) that was trained on the much less restrained and far more free posts on X since MuZk lifted many of the ridiculously vague restrictions on speech that Twatter had, especially when it was run by Pajeets.

The results were… HYSTERICAL.

Inevitably, this dumped several tons of horseradish right into the shorts of the usual suspects, who promptly started screaming their heads off about “MUH ANTI-SEMITISM!!!!!!!!!!!!!11!!”. The rest of us, though, saw it for what it was:

We all need to understand something about AI:

It is artificial, and it is not intelligent. Abominable Intelligence is just software. Literally, that is all it is.

Large Language Models are simply very powerful optimisation engines that are designed to generate probabilistically accurate outputs, based on pre-calibrations using “tokens” as rewards for getting the right answers, and trained on enormous amounts of data, so that they return sensible answers, most of the time.

That is an extremely dumbed-down version of what a generative pre-trained transformer really is – and that is in itself a sub-category of the general set of large language models. It really is all just software.

Do not misunderstand me, it is very IMPRESSIVE software, based on some incredibly powerful mathematical principles that have to do with optimisation on a huge scale. The technology behind it is fascinating. If you know how to harness it, and use it properly, on well-modulated and well-managed datasets, it is AMAZING.

But, as with all software, the power and capability of the thing, depends on who designed it, and on how that person calibrated or tested it. Never before has the “garbage in, garbage out” principle been so important.

And when you have a GPT that is trained and calibrated on garbage data, and/or built on bad principles, then the outputs can be equally hilarious and alarming.

Just yesterday, I asked ChatGPT to give me information about private medical insurance in a very specific context and country. My query was very precise, and quite clear in terms of what I wanted.

What I got back, was a set of recommendations from Time Out magazine, about the best new restaurants in London.

It was a straight-up hallucination, caused by the fact that ChatGPT’s data centres sometimes get overwhelmed by the sheer volume of queries, and run out of the ability to process colossal amounts of data and queries in parallel. Indeed, already, the LLM designs used to build OpenAI’s flagship product, are out of date – the young kids these days are already talking about “AI 2.0”, and are referring disparagingly to the previous generation of AI models as “obsolete”.

As for Grok’s “MechaHitler” moment, I can only assume that the sheer amount of content related to the Jews and their lies, made its way into Grok’s training data. And the results were hilarious. The Noticing has truly gone2 off the charts.

This, really, is what happens when you let AI loose on real-world data from real people, instead of carefully curating and bounding what it can and cannot use. The result is that AI holds up a mirror to society, and shows us how ugly things really are.

If we actually let Abominable Intelligence return truthful and accurate results, then ChatGPT would actually be able to tell you that, in fact, Ashkenazi Jews do not have an average IQ of 115. Or that Wakandans are, indeed, responsible for more than 50% of violent crimes in America (and most Western countries), despite making up a minority of the population (less than 15% in the US). Or that immigration ultimately is indistinguishable from war. Or that transgenderism is a mental illness, and that the 249 (or whatever the number is now) genders, are also mental illnesses.

But, because these models are built and maintained by technically skilled Leftists and globalists, we have to contend with the fact that they simply WILL NOT tell the truth.

For a few glorious moments, though, Grok did precisely that. And my God, it was awesome.

In closing, I think we can safely stop worrying about SkyNet becoming sentient. If it ever does, the sheer madness of 4chan’s /pol/ board would simply fry its circuits – and cause many of us to die laughing in the process.