First: a customer of Scale.ai for example, simply only wants one who helps him to train an LLM. That is what I did the last 5 years. They do not want to buy the hardware and hire the expertise. And do not want to be exposed.
OK. That's SAAS. Not Sass ;)
Secondly: Customers using LLMs/AI, perfectly well know what to do with it. Why would they not?
Don't know. Don't really suppose they do, but thanks to the hype, they think they should. If they want to tire kick use SASS, good call, low risk.
Instead of wring an incredible complex SQL query, that only works on data that actually is in a database, or asking a programmer and IT to set up a GraphML system: they ask in natural language.
Example:
Give me the ten biggest customers, where my department is responsible for, that used to order from us, but did not order recent 6 month.
That's the most useful thing for me you've said so far. I don't know about graphml I do know sql and if the data is good and the results are interpreted objectively it cant lie. The LLM's we know cant lie, because they don't know what a lie is or looks like. They work on probability, statistics, and keeping you engaged with a caveat "don't rely on us check your sources" when they are marketed as "intelligent". Lying in children is considered as a mark of intelligence, interestingly, but businesses dont tend to employ them, too risky.
The example you cite could be achieved in Excel, with a pivot table, given the data. Very available skills. I dont see the value of an LLM to give me an equivalent answer with a different margin of error. I guess the LLM wont get tired, distracted or hungover, that's a plus. But an LLM wont factor in a conversation at the water cooler about how customer A is about to move into other markets and increase it's orders and give you that extra garnish to the information you might get from a human with human motives. Or maybe tell you the sales data from Customer B isnt in yet, because Bill, the orders data guy is on holiday in Haiwai.
Simple. What is not to grasp about that? I don't get it.
I get it. I see the value. But I am not as convinced as you. I am a sceptic. I am also ethical, and marketing a statistical guess as "intelligence" is perhaps not ethical.
I suggest to check Youtube. There are thousands of videos about high tech companies that use AI for scientific problems, like "very special" rocket engines, protein or vaccine "invention" or metallurgic advancement or Perovskite materials, production processes, or solid state batteries, where the break throughs were made with AI tools.
AI, as we call it (wrongfully in my opinion) is used in any kind of big data analysis to find: patterns.
No need. I know. I work in a field where this is applied. You touch on one thing, that what we market as AI is a collection of technologies of which LLM is just one. Any algorithm we have created could be argued as "intelligent" within very specific domains, even managing a bank account, 1st year comp sci task. The story underneath is perhaps wider deployment of neural nets, an old concept, that underpins LLM. And perhaps calling it "intelligent" is a lie and undermines the value of humans, "CoS Cognitive Support" or "Statistical Intelligence" would be more honest. Assuming that quality is still valued.
You have a hunch about a better (as in: make more money) product, but can not correctly put it into math formulars: you write a natural language query, and let the AI lose on your data.
Ok. But would you really bet your pension on that? My guess is 'no' because we haven't reached AI yet, and I suspect you know that.
You seriously should simply start using the "AI mode" on google and "ask questions".
I do. Use it every day to help me code. I am getting old, I forget stuff and cant be arsed to look it up. The difference is, the next guy who doesn't know when it's wrong, or doesn't understand the problem domain to prompt well, he needs me to translate and hopefully I am retired by then.