I've not found even one good use case so far in my line of application engineering. What are people's opinions on how much of the current "AI will change the world like steam engine" sentiment is reality vs marketing?
2 replies (most recent on top)
@a9 Spot on. I feel the same frustration and have had to come to terms with AI getting slapped on everything that's automated to any degree. As long as I think "automation" the frustration evaporates. Really have to ask leaders/vendors/etc. "What type of AI?"
For LLMs, the large foundation models can't possibly find much more text to absorb. OTOH, enhancements (and efficiencies) to the # of hidden layers and memory size during training will continue to find new types of "links" in documented human efforts. For hard sciences, this may elucidate via hints on where humans might want to focus research. That alone can be a benefit, but yes...it won't do the work correctly for anyone!
On the other side of things, within hard science there are more and more use cases where we take what was learned in building LLMs and replace "human language" with molecules, physical attributes, equations, etc. to build custom-purpose models that only know a specific science (versus 'scooping some text about them up' for LLMs). That will produce enhancements for all of society, but we may not ever hear about them.
tldr; LLMs are a 'better than search' personal research tool...the important bit is now we know that it can be done...and should see incremental improvements in thousands of businesses/functions (once execs pull their heads out of their as--s)
Steam engine? No. That's hype.
But I expect it'll be more like the internet did. The internet required you to "sign up for it". The web started in 1994 or so, and took at least a decade for most people to get on it.
And we really should stop calling it AI, and start calling it an LLM, which is really what we're talking about. I've used it to generate some low-level code, or ramp up my knowledge base for something I don't know a lot about. It's a great learning tool. It's a decent research tool. It's not a great code generating tool, and it's an even poorer automation tool.
It's HUGELY incredible at finding very hard to find information. I can't tell you how useful it is at just scouring the web and finding and summarizing hard to find little bits of info. I can read a news article, figure out all the little bits about what's right and wrong about it, and what they're not really saying.
But it's not gonna cure cancer. write an application, or produce anything of value without a human being that's smarter than it checking its work. It's not going to "ki-l jobs", but it might create some new ones.