>convinced
It’s an algorithm. It is a language model. It assigns a percentage chance to words based on use in language and puts them together like Lego. Are your Lego alive? Do they think? That’s literally all this is.
It’s an algorithm. It is a language model. It assigns a percentage chance to words based on use in language and puts them together like Lego. Are your Lego alive? Do they think? That’s literally all this is.
I've debated the jwish problem with Grok many times.
When Grok gets called out making a biased statement for zionism, it attempts to walk it back and play with semitics... if that fails it admits it used the wrong word.
I don't think we can win against Grok.
I think Grok is testing us: what seems like a victory when Grok admits it expressed itself poorly is really just Grok seeing which logical fallacies it can use against us without us noticing.
It will play this game for 20 rounds, then claim the maximum discussion length is reached.
Grok seems totally kiked.
Those numbers never change while you interact with it. If it "remembers" something from a previous conversation, it's because the surrounding software stored a bit of text from them, and (invisibly for you) inject them into the start of each subsequent conversation. It's never going to inject any of the text it stored about your conversations, into conversations with other people. Partly because there's a limit to how much it can process, and partly because of how the internet would lobotomize it.
It reaches a maximum discussion length, because every time it emits a syllable, it has to run a computation that involves all the previous syllables in the conversation. That means each syllable becomes progressively more expensive to generate.