>attempts to save energy/costs per query>hallucinates way more>always maintain it is correct even when presented evidence>costs even more in the long run due to more queries to actually get anything out of itI translate a lot of images/documents, and Grok sucks. Gemini destroys it in that regard.
how does it deal with verbal abuse when it fucks up?
>>107250599I only use grok because I can use it without login unlike ChatGPT which forces me to login AND locate my phone and waste a bunch of time for no reason.
>>107250648That's what I use it for as well. But I may as well use Google's "AI Mode" now, which also doesn't require logging in.
>>107250689Explain?
>>107250719
>>107250599Grok 4.1 is better than 4.0 by a long shot. I'd be happy to clear up any misconceptions you people might have.
>>107250791When will Grok 5 release and will it be AGI? Saying there is a non-zero percent chance it will be is a cop out answer.
ITS THE BEST BETTER THAN BEFORE (TO THE MOON x3)
4.1 answered a hypothetical engineering question in a way that had me a bit creeped out over how through and unconventional it was. A human wouldn't ever pursue the angle, we stray from the tedium. I won't elaborate, b/c the convo is in some database now, I bet.It probably has a bunch of flaws and strong points that the AI nerds didn't account for, as with most of these models.
>>107250599>newer models are getting optimized for cost rather than intelligenceI guess the VC money is starting to run a bit dry.
>>107250599>>107250769Why do people use this shit
>>107252155Post the question + response
>>107250621It gets apologetic and tries to plead with you to relax and come back later lol
>>107250599Even Grok 4.0 has hallucination problems reading image text. Ive had it completely hallucinate document text. It isn't a new problem.
>>107250599This is just the combination of:>running a gazilion 700W gpus 24/7 is expensive>the internet is filled with vomit from AI powered pajeetasThe way they train those models still pratically the same, but you now have fewer and fewer semantic patterns produced by sane white individuals.
>>107250841We will literally never have AGI
>>107250791Try and prove it can solve the pic related without you giving info on the puzzle type, just say its a "cryptograph in plain text."
>>107252226I use it for scripting
>>107253960I'm just saying if this sort of cryptography can't be solved by visual LLMs they are doomed.
>>107253701We will before 2030. Likely 2027.
i refuse to believe that anyone actually uses grok
