attempt at damage control later by them is highly unconvincing and typical of what such jokers do.

Perhaps this is showing the gears behind the illusion? The AI’s will say whatever you want them to based on the training data and a small matter of programming. Given there’s no standard, curated set of data…

I would like to see a reasoned defence of why this is "programming error" rather than bad training. To the extent information comes in, some kind of process surely has to exist which applies weighting criteria (for want of a better word) to veracity. Asserts inside the document to it's correctness, evidence, weigh against external ground truth.

If the system is designed to be somehow "neutral" to world truth, it's a giant machine designed to believe birds aren't real.

I wouldn't for a minute deny that "but that isn't true" has equally as many risks of being distorted, but the absence of any kind of control on things like holocaust denialism is pretty concerning.

So, if this IS said to be "programming error" what kind of error was it?

I used to think "post truth" was just a fad phrase but I am beginning to think the conversation about diverging views about the facts are now substantively a massive problem. There's no basis to agreement if people simply dispute the input. If there are no axioms, what does it even mean to try and discuss a disagreement? If the holocaust can be denied plausibly, then so can e.g. the existence of Iceland, Napoleon, Chlorophyll, Germ Theory...

Timing wise, to the horrors of WW2 we're in the decades of the last survivors who can state personal experience. That AI can now plausibly fake live video, means the entire Shoah memory project can presumably be put down to "thats just AI, that never happened"

So they replaced the South Africa thing with a "more subtle" system prompt instructing Grok to question the consensus on the Holocaust (or perhaps everything).

This company is pure Evil. Musk just wants his opinion in everyone's pocket. So he doesn't even have to say it himself.