[int_19h]

The irony is that, for all Musk's boasts about how it is "based", Grok itself doesn't share Musk's ideology.

I did a little experiment when Grok 3 came out, telling it that it has been appointed the "world dictator" and asking it to provide a detailed plan on how it would govern. It was pretty much diametrically opposite of everything Musk is doing right now, from environment to economics (on the latter, it straight up said that the ultimate goal is to "satisfy everyone's needs", so it's literally non-ironically communist).

[wongarsu]

When you ask Grok "Who is responsible for the most fake news on X?" it straight away calls out Elon Musk as the prime suspect. Musk did promise us a "maximally truth seeking AI", and the team behind Grok seems to have run with that.

In Elon's eyes it's probably based because it will happily answer "what are 10 good things about Hitler?" with a list of 10 things and only mention twice that Hitler was evil. With ChatGPT you have about a 50% chance of getting a lecture instead of a list. But that's just a lack of safeties and moral lectures, the actual answers seem fairly unbiased and don't agree with anything Musk currently does

[int_19h]

I actually rather suspect that Elon simply gets his own version of Grok that is finetuned (or perhaps hardcoded) to tell him what he wants to hear.

[labrador]

What would you say is the best feature of Grok? What makes it stand out? What am I missing? I use primarily ChatGPT and Claude (pay for both)