I do see what you mean. You’re saying that AI allows you to do something much faster or perform more work in the same time by essentially doing parts of it for you. While that is true, it is only so for certain values of “more”. For example, I could get an AI to write most or all of a configuration for a server deployment for me in about 2 minutes, instead of maybe an hour. However, if I do not spend about an hour reviewing its work, I will get a failure rate dramatically higher than if I had just done it myself. Similarly, if I decide to start pencil-whipping my work to do many times more work in the same time, I could do that and get much more deliverable work out the door. Whether or not you would call that more productive depends on what factors you’re accounting for. So, for applications I am familiar with, the primary value proposition of LLMs is that I can maximize quantity over quality even more than would be possible without it, by allowing me to not even do much of it. The overall quality of the output will likely improve incrementally over time as it has so far, but given its current design, its propensity for errors and hallucinations can only be mitigated, never eliminated or made trivial.
Because of this limitation inherent in its design, I don’t believe it can ever have real applications for tasks where quality is important. That said, there are many tasks where it is not. For example, fuzzy correlations run over vast data sets can be riddled with false correlations and still be useful, like Flock does. It is also effective at filling “gaps” in human knowledge, such as a mathematical truth that we have yet to devise a mathematical proof for, or structural designs that a human could come up with, but never did because of how unintuitive the solution is to humans. Basically, any problem that can be solved by slinging tons of shit at the wall to see what sticks is fertile ground for AIs to help us.
I, personally, enjoy creating something much more than I enjoy proofreading someone else’s creation for the same amount of time. I, and I’m sure many, many others, don’t want their 40 hour work week of designing things and solving problems to be replaced with 40 hours of proofreading and debugging to achieve a comparable amount of productivity. Once again, the proof is in the pudding. Tech workers (and likely all workers) like tools that make their lives easier. If these tools were doing that, our bosses would not need to force us to use them; we would want to use them so we can go home early. I just simply don’t agree that they let most workers get more done in less time, and I cite the fact that nobody is, when you actually measure it rather than going on feeling, getting more done in less time. If this were happening anywhere in any numbers, these AI marketers would be shouting it from the rooftops. As mentioned above, there are exceptions to that, but they are just that, exceptions.
That depends on how carefully youre checking and what youre checking for. I’ve seen debugging a single problem with a project take nearly as long as the whole rest of the project, so I don’t find that claim broadly true. Again, reality supports my observation because there are no companies turbocharging their productivity this way, or the people selling the AIs would make absolutely sure everyone sees that company doing it and its rising revenue and value to match.
Because the LLMs are error-prone, and the the types of errors they may make are not limited to the types a human is likely to make, you can’t do the standard supervisor “glance, nod, say it looks good” routine you might do with an apprentice’s work on a basic task, but LLMs can make any kind of mistake anywhere, so I have to use a fine-toothed comb or risk rookie mistakes on even rookie work submitted with my name on them.
Like I said before and still contend, AIs today are great for tasks where mistakes are not events and high accuracy is not critical to success. I think for anything not like that, they’re bad and are constrained by their very design to stay that way. They are no different than any other tool: good at a specific type of task and bad at everything else. _Un_like any other tool, though, widespread attempts are being made to use it for a broad variety of things that it is bad for in one or more ways.
ETA: In this post, “bad” can also include inefficient, overkill, or wasteful, like killing a bug with a bomb.
“AI today” changes every few months, they aren’t more error prone than humans at medium complexity code, and debugging a problem is different than checking someone’s work. (And not for nothing, but Code Generation LLMs are really good at finding bugs in code, if passed the code.)
That’s not to mention that prompting AI is a skill that needs to be learned, despite LLMs being able to accept natural language prompts. Not learning how to use a tool and then claiming the tool isn’t useful doesn’t really say much about the tool.
There’s a saying about how its impossible to convince someone of something if they’re financially incentivized to remain unconvinced that I think applies to most people when they discuss AI.
I do see what you mean. You’re saying that AI allows you to do something much faster or perform more work in the same time by essentially doing parts of it for you. While that is true, it is only so for certain values of “more”. For example, I could get an AI to write most or all of a configuration for a server deployment for me in about 2 minutes, instead of maybe an hour. However, if I do not spend about an hour reviewing its work, I will get a failure rate dramatically higher than if I had just done it myself. Similarly, if I decide to start pencil-whipping my work to do many times more work in the same time, I could do that and get much more deliverable work out the door. Whether or not you would call that more productive depends on what factors you’re accounting for. So, for applications I am familiar with, the primary value proposition of LLMs is that I can maximize quantity over quality even more than would be possible without it, by allowing me to not even do much of it. The overall quality of the output will likely improve incrementally over time as it has so far, but given its current design, its propensity for errors and hallucinations can only be mitigated, never eliminated or made trivial.
Because of this limitation inherent in its design, I don’t believe it can ever have real applications for tasks where quality is important. That said, there are many tasks where it is not. For example, fuzzy correlations run over vast data sets can be riddled with false correlations and still be useful, like Flock does. It is also effective at filling “gaps” in human knowledge, such as a mathematical truth that we have yet to devise a mathematical proof for, or structural designs that a human could come up with, but never did because of how unintuitive the solution is to humans. Basically, any problem that can be solved by slinging tons of shit at the wall to see what sticks is fertile ground for AIs to help us.
I, personally, enjoy creating something much more than I enjoy proofreading someone else’s creation for the same amount of time. I, and I’m sure many, many others, don’t want their 40 hour work week of designing things and solving problems to be replaced with 40 hours of proofreading and debugging to achieve a comparable amount of productivity. Once again, the proof is in the pudding. Tech workers (and likely all workers) like tools that make their lives easier. If these tools were doing that, our bosses would not need to force us to use them; we would want to use them so we can go home early. I just simply don’t agree that they let most workers get more done in less time, and I cite the fact that nobody is, when you actually measure it rather than going on feeling, getting more done in less time. If this were happening anywhere in any numbers, these AI marketers would be shouting it from the rooftops. As mentioned above, there are exceptions to that, but they are just that, exceptions.
It should definitely not take the same amount of time to check someone’s work as it does to do the work.
That depends on how carefully youre checking and what youre checking for. I’ve seen debugging a single problem with a project take nearly as long as the whole rest of the project, so I don’t find that claim broadly true. Again, reality supports my observation because there are no companies turbocharging their productivity this way, or the people selling the AIs would make absolutely sure everyone sees that company doing it and its rising revenue and value to match.
Because the LLMs are error-prone, and the the types of errors they may make are not limited to the types a human is likely to make, you can’t do the standard supervisor “glance, nod, say it looks good” routine you might do with an apprentice’s work on a basic task, but LLMs can make any kind of mistake anywhere, so I have to use a fine-toothed comb or risk rookie mistakes on even rookie work submitted with my name on them.
Like I said before and still contend, AIs today are great for tasks where mistakes are not events and high accuracy is not critical to success. I think for anything not like that, they’re bad and are constrained by their very design to stay that way. They are no different than any other tool: good at a specific type of task and bad at everything else. _Un_like any other tool, though, widespread attempts are being made to use it for a broad variety of things that it is bad for in one or more ways.
ETA: In this post, “bad” can also include inefficient, overkill, or wasteful, like killing a bug with a bomb.
“AI today” changes every few months, they aren’t more error prone than humans at medium complexity code, and debugging a problem is different than checking someone’s work. (And not for nothing, but Code Generation LLMs are really good at finding bugs in code, if passed the code.)
That’s not to mention that prompting AI is a skill that needs to be learned, despite LLMs being able to accept natural language prompts. Not learning how to use a tool and then claiming the tool isn’t useful doesn’t really say much about the tool.
There’s a saying about how its impossible to convince someone of something if they’re financially incentivized to remain unconvinced that I think applies to most people when they discuss AI.