its one of the use-cases that AI truly makes sense in to me, because it feels like voice assistant technology has really plateau’d, and an LLM seems like a good way to process natural language
I somewhat bought into the hype early and convinced work to pay for ChatGPT plus. At first I struggled to use it. One day I somewhat went “I bet it can’t help with X”, it did. Now I’m at the point where I default to it. There is this odd assumption that it will only be right some of the time. To me it’s rare where it’s wrong. Usually it mainly misunderstood the direction I was trying to go in and once I fix it with follow-up prompt I get what I want.
I don’t think I do prompt engineering per se. It’s like google fu though. You need to learn to be descriptive to the point where the LLM can infer some context then even a year later it feels surreal. So far GPT-4 is the top for me. llama does well and a lot of the open models are nice. But if I want code or think through some work problem, GPT-4 gets me where I want to get amazingly fast. I make it do online research for me and then I have it validate my thoughts. I have to keep in mind “hey, it’s mainly predicting the next word”. But I rarely go “wow it was truly off here”. Trust but verify is where I’m at.
I’m at the point where I feel like I do my 40 hour work week in 25 or so. I have a ton more free time. I have to be careful not to share any direct work related info, but that’s easy. I give it generic info then fill in the blanks myself.
I make it do online research for me and then I have it validate my thoughts.
That’s precisely the issue. The words sound convincing, but this way of thinking leads to it becoming a yes-man. Either it confirms what you think, or your prompt is wrong.
Honestly, I confirm it because I use it for work. I had it do some research on comparing bunch of VDI solutions (the VMware/Broadcom thing has forced us to rethink things). It did a really good job summarizing things. I used to work in consulting, so I already knew what the comparison. It saved me hours of having to write that report. I usually verify in the term that “does it make sense”. I would do the same with a stackoverflow post before posting the code and so on.
its one of the use-cases that AI truly makes sense in to me, because it feels like voice assistant technology has really plateau’d, and an LLM seems like a good way to process natural language
I somewhat bought into the hype early and convinced work to pay for ChatGPT plus. At first I struggled to use it. One day I somewhat went “I bet it can’t help with X”, it did. Now I’m at the point where I default to it. There is this odd assumption that it will only be right some of the time. To me it’s rare where it’s wrong. Usually it mainly misunderstood the direction I was trying to go in and once I fix it with follow-up prompt I get what I want.
I don’t think I do prompt engineering per se. It’s like google fu though. You need to learn to be descriptive to the point where the LLM can infer some context then even a year later it feels surreal. So far GPT-4 is the top for me. llama does well and a lot of the open models are nice. But if I want code or think through some work problem, GPT-4 gets me where I want to get amazingly fast. I make it do online research for me and then I have it validate my thoughts. I have to keep in mind “hey, it’s mainly predicting the next word”. But I rarely go “wow it was truly off here”. Trust but verify is where I’m at.
I’m at the point where I feel like I do my 40 hour work week in 25 or so. I have a ton more free time. I have to be careful not to share any direct work related info, but that’s easy. I give it generic info then fill in the blanks myself.
That’s precisely the issue. The words sound convincing, but this way of thinking leads to it becoming a yes-man. Either it confirms what you think, or your prompt is wrong.
Honestly, I confirm it because I use it for work. I had it do some research on comparing bunch of VDI solutions (the VMware/Broadcom thing has forced us to rethink things). It did a really good job summarizing things. I used to work in consulting, so I already knew what the comparison. It saved me hours of having to write that report. I usually verify in the term that “does it make sense”. I would do the same with a stackoverflow post before posting the code and so on.