Thanks for the thoughts, appreciated. However, you're both missing a key point: generative Ai simply isn't. Rather, it's generally plagiaristic and deriviativistic because its training is based entirely on all the data stolen globally from sources from all across the internet.
True, but the researchers have shown that this tremendous amount of information is enough---and with the LLM algorithms---for them to actually discern some kinds of fairly sophisticated reasoning and partial theory of mind. Because all of that was implicitly encoded in the corpus of the training text, and at last the algorithms & data processing is good enough to figure it out. Plagiaristic and derivative, sure but so are most humans.
The success of these models was a surprise to the researchers (and myself)---there is nothing built into them that seems like it would make them that intelligent, no tremendous deep breakthroughs. The transformer architecture is clever and computationally efficient but not all that amazing. R&D is now back engineering some of the processes to figure out what they are doing inside---synthetic neuroscience.
Given how poorly the muskrat generally handles software development (e.g. FSD, Automatic Wipers, Tesla Vision, Twitter, et. al.), there's no way Grok (it's name is yet another theft, this time from a "content creator" named Robert Heinlein) will be any better than the others, and this one could easily be far worse. So, while I'll watch this creation with interest, I'll not give an inch on the proposition that it is not now, nor will it likely ever be intelligent.
Their researchers are good, but more likely Elon will lose interest soon (which will improve their performance if he leaves them alone) but then later cease funding as there is no good business model. Anyway, it's picking Elon's pocket provides potentially some good, if at minimum, it provides effective competition to OpenAI, which is the opposite of its name, now entirely LockedDownSecretAI. Surprisingly, it's Facebook/Meta, led by Yann LeCun, which is the most deeply open of any major corporate AI lab these days. Zuck isn't the bad guy for once.
I am not saying that AI bots cannot be useful - quite the contrary. However, these collections of software should not be referred to as AI, because they simply aren't. Sometimes useful, interesting, capable of producing surprising and helpful responses, in most cases easily better than standard Internet searches, yes, but not intelligent, They may even be uniquely helpful in realizing the dream of FSD and autonomous driving, but right now every concept or conclusion they produce must be carefully checked before it's adopted and put to use because it may be a lie, it may be a confabulation, it may be highly biased, it may be plagiarized, and so on.
This is an important point and one, remarkably, entirely unanticipated by science fiction as far as I know. Sci-fi typically assumed that the computer's superior performance and reliability on rule-based computational tasks would carry over to future AI models, with humans still superior for creativity. But it's turning out that reliability and factuality are the major unsolved problems and you need humans to check their output, while the AI's are 'creative' (within genres, not fully genre-breaking like actual human geniuses). Technically, the LLMs are trained by predicting the next symbol forward and they don't have a strong notion of true that's different from 'highly probable in my train corpus', something that even small children understand well.
Oh, and it's my understanding that as of now, nothing that comes out of a large language model can be coprighted, so that's actually kinda cool. As a result of this, it appears that anything one of these engines produces would fall directly and immediately into the public domain - that list would necessarily include art, text, computer code, and so on.
Probably not---if it's considered like an artist's paintbrush legally, then the wielder of the brush maintains legal right to the production. Or likewise if it's a work-for-hire like most employer-employee relationships then the owner also gets the copyright.
www.teslarati.com
Tesla cars will run a small version of Grok some year.
