Except like... it's not solving that either.
A metaphor I've been using lately is that being a tech-interested person and watching the AI hype is like if you had followed the development of blenders for years. You watched them go from prototypes that were basically just a spinning open blade all the way to a design that has the potential to be a consumer Vitamix! It's really cool! Blenders have come such a long way, and they're ready for prime time!
And then you turn on the news and see otherwise rational, intelligent people saying "gosh, imagine, soon we'll replace all of our chefs, and our surgeons, and our high school teachers with blenders!" and your friends and family all nod and agree and say things like "wow, blenders can basically do everything now!" And when you ask people "are you HIGH?!" they show you the new blender they bought, and how well it makes a smoothie, and then they act like that's evidence for a statement like "blenders will replace 90% of the workforce" not being utterly nonsensical and deranged.
"Scientists just have to fix the hallucination problem!" they say. When you ask what the hallucination problem is, they say "I'll show you" and then they put their unfinished math homework into the blender and hit pulse. "You see, I wanted it to solve those math problems, but it just shredded the paper. It must have hallucinated a world where the answer to 2+2 was puree." When you point out that it did exactly what it was designed to do, because a blender cannot do math and it will never do math and expecting it to be able to do math just because it can make both smoothies and soup is ludicrous and bizarre, they tell you that they're sure that blenders will be able to do math any day now, just you wait, "I mean, look how far they've come! A year ago I would have said that blenders could never be strong enough to blend ice into sorbet, but now they can. So who are you to say they'll never do math???"
The thing is, there are plenty of things that LLMs and generative AI are good for. OCR is still a vital need, and AI is excellent at it. Facial recognition is an area where AI has a lot of potential. It can be used as a screening step for the analysis of all sorts of large datasets. Better autocomplete on your phone is a real thing that real people want. There are a ton of problems that these tools genuinely do solve!
...But none of them are trillion dollar problems, and that's an issue, because no one wanted to pour a trillion dollars into "improved OCR" and "somewhat better autocomplete".
So, we get snakeoil to make up the weight. It'll solve payroll; it'll democratize visual art; it'll make it so that anyone can do anything – you name it, blenders will do it, eventually! Once we've fixed the problems with using blenders to do everything, including tasks which aren't blending things, then blenders will be worth the trillion dollars that people have already spent developing them! Look at the progress we've already made: we're working on attaching a calculator to the blender, and that could make it so that blenders can do math! Don't you dare suggest that calculators already exist and work just fine without being attached to blenders – we need our blenders to do math, because we promised that one day they'd be able to do everything! And they will! Blenders are the future and don't you dare suggest that maybe that future is just "we can blend things now"; if you so much as breath those words, the bubble will pop.
