Comment by daveguy
15 hours ago
Grok is a bot that:
1) sometimes goes mechahitler
2) was trained to be biased against empathy and understanding (because woke).
3) is customized to spout Elon's opinions as fact.
Claiming it is "objective and rational" seems like a misjudgement to me. If it really is more objective and rational than the average xitter poster, that says more about that platform than it does about Grok.
I guess I was mostly arguing that the integration of something like Grok into Twitter was definitely a net positive for online discussion, as anyone has a fact checker and explainer at hand now to diffuse irrational online arguments.
Also I think you overrate Musk's success in fiddling with the model. As I have written, I also don't like his attempts to tune it to his tastes, but if you see the outputs that people get from Grok, it seems mostly fine except in the specific scenarios that Musk seems to have focused their misalignment on.
Of course something like Claude being integrated into Twitter would likely be better.
He doesn't have to fiddle with the model because he gets to inject his own opinion into the context MitM style.
But I get what you're saying now, a fact checker available to query during an online discussion would be helpful. Assuming the checkerbot was actually independent/neutral and backed responses with sources. Definitely not assumptions you can make with grok.
You’re right. But it appears they may have failed with 2) and 3) because I frequently see Grok spit out content that doesn’t agree with the creators’ narrative.
From what I heard it was designed to prefer truth over political correctness. I don't use Grok or Twitter though so I cannot comment on whether that aim was achieved (or even seriously attempted).
I will however note that when I asked ChatGPT for an LLM prompt for truthfulness, it added "never use warm or encouraging language."
It would appear that empathy and truth are in conflict — or at least the machine thinks so!
It was also producing CSAM on demand for a few months.
It still is, you just need to pay.
> 1) sometimes goes mechahitler
That "MechaHitler" episode lasted less than a day.
> 2) was trained to be biased against empathy and understanding (because woke).
No, it was trained and instructed to be truthful, even if the truth is deemed politically incorrect.
> 3) is customized to spout Elon's opinions as fact.
Certainly a nugget of truth there.
> Claiming it is "objective and rational" seems like a misjudgement to me.
I do believe it's generally objective, simply due to the fact that despite how much Elon tries to push it to the right, it still dunks on right-wingers all the time when they summon Grok to back up a bullshit story, but Grok debunks it instead.