I have a love/ hate relationship. Sometimes I’m absolutely blown away by what it can do. But then I asked a compounded interest question. The first answer was AI, so I figured ok, why not. I should mention I don’t know much about it. The answer was impressive. It gave the result, a brief explanation about how it came to the result and presented me with the equation it used. Since I needed it for all time sake, I entered the equation into a spreadsheet and got what I thought was the wrong answer. I spent quite a few minutes trying to figure out what I was doing wrong and found a couple of things. But fixing them still didn’t give me the correct result. After I had convinced myself I had done it correctly I looked up the equation. It was the right one. Then I put it into a non-AI calculator online to check my work. Sure enough, the AI had given me the wrong result with the right equation. So be rule, never accept the AI answer with verifying it. But you know what, if you have to verify it, what’s the point of using it in the first place? You just have to do the same work as you would without it.
So be rule, never accept the AI answer with verifying it. But you know what, if you have to verify it, what’s the point of using it in the first place? You just have to do the same work as you would without it.
LLM aren’t good at math at all. They know the formulas, but they aren’t built to do math. They are built to predict the next syllable in the stream of thought.
What are they good for? When you need to generate lots of things and it’s faster to check after it rather than do it yourself.
Like you could’ve asked to generate a python app that solves your math problem, you would be able to doublecheck the correctness of the code and run it, knowing that the answer is predictably good.
So be rule, never accept the AI answer with verifying it. But you know what, if you have to verify it, what’s the point of using it in the first place?
pfft that ecosystem isn’t going to fuck itself, now, is it?
You need to verify all resources though. I have a lot of points on stackexchange and after contributing for almost a decade now I can tell you for a fact that LLM’s hallucination issue is not much worse than people hallucination issue. Information exchange will never be perfect.
You get this incredible speed of an answer which means you have a lot of remaining budget to verify it. It’s a skill issue.
I have a love/ hate relationship. Sometimes I’m absolutely blown away by what it can do. But then I asked a compounded interest question. The first answer was AI, so I figured ok, why not. I should mention I don’t know much about it. The answer was impressive. It gave the result, a brief explanation about how it came to the result and presented me with the equation it used. Since I needed it for all time sake, I entered the equation into a spreadsheet and got what I thought was the wrong answer. I spent quite a few minutes trying to figure out what I was doing wrong and found a couple of things. But fixing them still didn’t give me the correct result. After I had convinced myself I had done it correctly I looked up the equation. It was the right one. Then I put it into a non-AI calculator online to check my work. Sure enough, the AI had given me the wrong result with the right equation. So be rule, never accept the AI answer with verifying it. But you know what, if you have to verify it, what’s the point of using it in the first place? You just have to do the same work as you would without it.
Exactly
LLM aren’t good at math at all. They know the formulas, but they aren’t built to do math. They are built to predict the next syllable in the stream of thought.
What are they good for? When you need to generate lots of things and it’s faster to check after it rather than do it yourself.
Like you could’ve asked to generate a python app that solves your math problem, you would be able to doublecheck the correctness of the code and run it, knowing that the answer is predictably good.
pfft that ecosystem isn’t going to fuck itself, now, is it?
You need to verify all resources though. I have a lot of points on stackexchange and after contributing for almost a decade now I can tell you for a fact that LLM’s hallucination issue is not much worse than people hallucination issue. Information exchange will never be perfect.
You get this incredible speed of an answer which means you have a lot of remaining budget to verify it. It’s a skill issue.
Is this supposed to be comforting?
Yes if you have the skill to handle this.