Grok by XAI – Elon Musk(x.ai)
x.ai
Grok by XAI – Elon Musk
https://x.ai/model-card/
20 comments
There’s no details here about training, evaluation, reproducibility or even architecture.
OpenAI’s GPT4 tech report was a marketing press release in disguise; I felt it represented a huge step down for the field in terms of academic rigour. But this model card epitomizes the end conclusion of LLM commoditization: no attempt to quantify limitations or weaknesses, portraying model biases and hateful/hallucinatory output as a feature rather than a bug, all wrapped up in breathless language of wanting to “maximally benefit humanity” while hiding the model behind a secretive waitlist or a $16 Twitter premium subscription plan.
What a boring release, to be honest.
OpenAI’s GPT4 tech report was a marketing press release in disguise; I felt it represented a huge step down for the field in terms of academic rigour. But this model card epitomizes the end conclusion of LLM commoditization: no attempt to quantify limitations or weaknesses, portraying model biases and hateful/hallucinatory output as a feature rather than a bug, all wrapped up in breathless language of wanting to “maximally benefit humanity” while hiding the model behind a secretive waitlist or a $16 Twitter premium subscription plan.
What a boring release, to be honest.
Elon says[1] the AI assistant will be part of X premium for $16/month.
More importantly:
>Grok is an AI modeled after the Hitchhiker’s Guide to the Galaxy, so intended to answer almost anything and, far harder, even suggest what questions to ask!
I'm a big fan of both, but using a Robert Heinlein word to name something you say is based on a Douglas Adams idea is weird.
1: https://twitter.com/elonmusk/status/1720839331365929290
More importantly:
>Grok is an AI modeled after the Hitchhiker’s Guide to the Galaxy, so intended to answer almost anything and, far harder, even suggest what questions to ask!
I'm a big fan of both, but using a Robert Heinlein word to name something you say is based on a Douglas Adams idea is weird.
1: https://twitter.com/elonmusk/status/1720839331365929290
He's that guy in your group who makes references to things he doesn't actually understand. This is the same guy that said the Cybertruck is what "Bladerunner" would drive. There is no such character named Bladerunner, his name is Deckard.
Ahem..his name is Billy.
https://en.wikipedia.org/wiki/The_Bladerunner
https://en.wikipedia.org/wiki/The_Bladerunner
> I'm a big fan of both, but using a Robert Heinlein word to name something you say is based on a Douglas Adams idea is weird.
This doesn't have anything to do with your point. It's just that I see people call things "weird" all the time on this forum. The way you've framed it comes off as a criticism, but is something being weird a bad thing?
This doesn't have anything to do with your point. It's just that I see people call things "weird" all the time on this forum. The way you've framed it comes off as a criticism, but is something being weird a bad thing?
If it doesn't make sense, then ... yes.
Hallucinating what questions to ask is far easier than answering correctly and definitively in my opinion.
This is pretty huge considering what it will have behind it. Namely a lot of Tesla's proprietary AI hardware and software.
Consider how good Elon is at pulling in talent and reducing complexity. Very exciting stuff.
Consider how good Elon is at pulling in talent and reducing complexity. Very exciting stuff.
No paper and not open source (yet?). I wonder how this is compatible with the lofty goals of furthering research and 'maximally benefit all of humanity'.
Musk was sharing some comments on twitter, praising its sarcasm. But isn't this type of trickery what AI protections should guard against?
Seems to be GPT-3.5 level. Not sure why anyone would pay for that given ChatGPT has a free plan, Claude, Bard, and the whole world of open source models.
Is it flagged by mistake ?
A high schooler could put together a more credible model card that this one.
superchroma(1)
But wryness aside, having it present itself to the user up front as potentially misleading, via humour, is probably the best way to operate over human minds anyway. Wit engages your own wit and critical faculties. If you are funnier than this AI, you are exercising your intelligence. Edward Tufte's old "the cognitive style of powerpoint" essay was about how the tech Microsoft used created generations of drones who thought in cliches and bullet points, whereas if Grok can actually challenge people via wit, we could actually get some enhancing effects from it. Psyched for this anyway. Congrats.