Stochastic Parrots Will Never Be AI
2024-11-04
It seems the wave of hype surrounding LLMs and associated technology has finally peaked. As all the mega tech monopolies raced each other to buy more GPU’s and plagiarize ever more creative works, it seems they didn't really worry about whether it would actually help them to sell products to real end users.
Many smaller businesses got caught up in the hype cycle as well, forcing LLM's into user experiences where accuracy mattered.
Apple Intelligence is the event we needed to assess what big tech really think of their AI creations. It turns out it's so mediocre, Tim didn't even turn up to talk about their new Mac line - “With Apple Intelligence”.
Breaking with tradition from the past few years the product lines received some spec bumps without increasing in price! The same with the iPhone launch in September… With nothing helpful to sell Apple only vaguely suggests that the AI features are valuable to users, just not valuable enough they would consider paying actual money for them.
The reason for this post is to encourage you to reassess if your ongoing LLM related projects are really going to deliver tangible benefits to your business, and if you have hastily added AI features to your existing software, there may be no better time than now to re evaluate if they truly need to be there.
The No Sweat video at the end of the MacBook Pro announcement was a real demonstration of the technology, I suppose accidentally from apples part, and it was a sight to behold. While this video was using normal special effects techniques rather than generative video models it thoroughly demonstrates the non existent use cases of the current crop of stochastic parrots.
Apple showing nonsense, because they have no valuable features.
Apple - The second most valuable company in the world currently as of this writing, has just been overtaken again by NVIDIA - could only come up with a weightlifter performing nonsensical impossible feats as the best way they could describe the incredible power of M4 - with Apple Intelligence. That's it, the best they’ve got. They should have ideally just gone with Shrimp Jesus I think. A live demo of how many Shrimp Jesus pictures apple M4 can generate and upload to Facebook would at least be a real world use case.
Generative models have one purpose - to churn out semi-believable gibberish, and this is why no one should take anyone forcing it into products seriously. You owe your users the best experience, whether that is a person reading a post on your website, using your application, or contacting you for customer support. When those users interact with a generative model you have signalled to them that you do not care about their experience at all.
During the presentation Apple also showed a "researcher" using summarization to draft their abstract and generating an image to include within their research paper, another fantastic use case of this amazingly useless technology.
Apple think researchers should be adding generated images in their scientific papers.
I'm not saying that LLM's couldn't have generated this post, and done it better & faster than me with no spelling mistakes, but they can't discuss this apple event for another year or two until they have 'been trained on' stolen the hard work of real people who are, occasionally, intelligent.