LLMs are a tool just like any other. There are tasks they can be useful, and others that they’re not.
The problem is that corporations and others are trying to force it down our throats in every single place as possible in their maddening quest to get a return in their investments.
Kinda sounds like most tools. If I try to fry an egg using a screwdriver its going to go very poorly for the egg and my cookware.
That’s the main issue with AI. It’s use cases have not been properly identified by the population. Unfortunately what it does wrong is not as obvious as my example. We built the tool and started using it before we knew how to use it.
Well that’s ridiculous, there are tons of tools that require months of training to use properly. You may have a general idea what a tool is supposed to do, but actually doing it is different.
I couldn’t hand you an electricians toolkit and tell you to rewire a house. And if I wanted you to change the brakes on a car, but I gave you a surgeon’s kit you would never be able to do it.
Tools have specific use cases, and you need to know when and how to use them to be effective. One of my biggest gripes about AI is this idea that it will provide. Nope it is 100% reliant on human knowledge, and I agree with the poster above processing specific datasets based on your specific instructions is about the only good use among the general public.
However specifically trained models have been assisting in some cool stuff. Archaeologists are using AI tools to “unravel” carbonized scrolls from Pompeii. A human alone could not do this. A public facing AI/LLM model could not do this. But a educated human training the model and overseeing it’s functions can. That’s tool use right there
Respectfully disagree. It’s because of the systematic insistence that we need to build data centers to support a demand that doesn’t exist.
Same thing would happen if corporations lost their minds over hammers and insisted we had to build hammer factories everywhere and sell hammers to stirr your coffee.
It’s rare that “not very accurate” and “expensive” are both acceptable together in a tool. They fail on two axes of the fast/good/cheap spectrum so I still think it’s an extremely niche tool, unlike a hammer which still fulfills an extremely common need for relatively little money.
Being common or niche has nothing to do with the quality of the tool. In fact LLMs should be a niche tool, because there’s only a few use cases where they are genuinely useful.
Thus what I said before, the problem is the companies trying to shove it everywhere without thought or care.
On being accurate or expensive, it depends heavily on what model we’re talking about and what exactly you’re using it for.
Put a light model summarising your web search, yeah, it’s an accident waiting to happen. Put a coding model helping you out building a small function while you focus on the main algorithm, it will save you a lot of time.
LLMs are a tool just like any other. There are tasks they can be useful, and others that they’re not.
The problem is that corporations and others are trying to force it down our throats in every single place as possible in their maddening quest to get a return in their investments.
They are a tool but they aren’t really like any other. They’re only rarely useful and have massive negative externalities.
Kinda sounds like most tools. If I try to fry an egg using a screwdriver its going to go very poorly for the egg and my cookware.
That’s the main issue with AI. It’s use cases have not been properly identified by the population. Unfortunately what it does wrong is not as obvious as my example. We built the tool and started using it before we knew how to use it.
Most tools do have obvious uses, like screwdrivers.
Well that’s ridiculous, there are tons of tools that require months of training to use properly. You may have a general idea what a tool is supposed to do, but actually doing it is different.
I couldn’t hand you an electricians toolkit and tell you to rewire a house. And if I wanted you to change the brakes on a car, but I gave you a surgeon’s kit you would never be able to do it.
Tools have specific use cases, and you need to know when and how to use them to be effective. One of my biggest gripes about AI is this idea that it will provide. Nope it is 100% reliant on human knowledge, and I agree with the poster above processing specific datasets based on your specific instructions is about the only good use among the general public.
However specifically trained models have been assisting in some cool stuff. Archaeologists are using AI tools to “unravel” carbonized scrolls from Pompeii. A human alone could not do this. A public facing AI/LLM model could not do this. But a educated human training the model and overseeing it’s functions can. That’s tool use right there
Respectfully disagree. It’s because of the systematic insistence that we need to build data centers to support a demand that doesn’t exist.
Same thing would happen if corporations lost their minds over hammers and insisted we had to build hammer factories everywhere and sell hammers to stirr your coffee.
It’s rare that “not very accurate” and “expensive” are both acceptable together in a tool. They fail on two axes of the fast/good/cheap spectrum so I still think it’s an extremely niche tool, unlike a hammer which still fulfills an extremely common need for relatively little money.
Being common or niche has nothing to do with the quality of the tool. In fact LLMs should be a niche tool, because there’s only a few use cases where they are genuinely useful.
Thus what I said before, the problem is the companies trying to shove it everywhere without thought or care.
On being accurate or expensive, it depends heavily on what model we’re talking about and what exactly you’re using it for.
Put a light model summarising your web search, yeah, it’s an accident waiting to happen. Put a coding model helping you out building a small function while you focus on the main algorithm, it will save you a lot of time.