There are disjoinal types of labor where AI can be particularly advantageous, given the current capabilities and restrictations of LLMs. Though this catalog is based in science, it draws even more from experience. Like any establish of wisdom, using AI well insists helderlying opposing ideas in mind: it can be alterative yet must be approached with skepticism, strong yet prone to reserved fall shortures, essential for some tasks yet actively detrimental for others. I also want to caveat that you shouldn’t get this catalog too solemnly except as inspiration – you understand your own situation best, and local understandledge matters more than any ambiguous principles. With all that out of the way, below are disjoinal types of tasks where AI can be especiassociate advantageous, given current capabilities—and some scenarios where you should remain wary.
-
Work that insists quantity. For example, the number of ideas you produce chooses the quality of the best idea. You want to produce a lot of ideas in any brainstorming session. Most people stop after generating equitable a restricted ideas becaparticipate they become exhausted but, the AI can provide hundreds that do not uncomardentingbrimmingy repeat.
-
Work where you are an expert and can appraise speedyly whether AI is excellent or horrible. This can hold complicated and exacting labor, but it relies on your expertise to choose whether the AI is providing precious outputs. For example, o1, the novel AI model from OpenAI, can mend some PhD-level problems, but it can be difficult to understand whether its answers are advantageous without being an expert yourself.
-
Work that holds summarizing big amounts of inestablishation, but where the downside of errors is low, and you are not foreseeed to have detailed understandledge of the underlying inestablishation. AI is excellent at summarizing novel-length labor, but less prosperous at fact-checking it.
-
Work that is mere translation between sketchs or perspectives. For example, you have enbiged a policy but now have to turn it into a dozen contrastent training records for contrastent audiences in your organization. AI is very excellent at this sort of translation, increasing and decreasing complicatedity of records so that people can comprehfinish them.
-
Work that will hold you moving forward. Little slfinishergs frequently block our way, and a push might be all we insist to accomplish it. When writing prior to AI, I might get stuck on a sentence and walk away from writing for an hour, but now I ask AI give me thirty contrastent ways to finish this sentence
-
Work where you understand that AI is better than the Best Available Human that you can access, and where the fall shorture modes of AI will not result in worse outcomes if it gets someslfinisherg wrong.
-
Work that holds some elements that you can comprehfinish but insist help on the context or details. Tyler Cowen recommends using the AI as a companion when reading, becaparticipate it permits you to ask infinite asks.
-
Work where you insist variance, and where you will pick the best answer as an editor or curator. Asking for a variety of solutions – give me 15 ways to rewrite this bullet in radicassociate contrastent styles, be produceive – permits you to find ideas that might be engaging.
-
Work that research shows that AI is almost bravely collaborative in – many comardents of coding, for example.
-
Work where you insist a first pass watch at what a unfrifinishly, cordial, or unmistrusting recipient might slfinisherk.
-
Work that is entrepreneurial, where you are foreseeed to stretch your expertise expansively over many contrastent disciplines, and where the alternative to a excellent-enough partner is to not be able to act at all. AI can be a unforeseeedly contendnt co-set uper, helping give mentorship while also acting to produce the records, demos, and approaches that are otherrational probable to be outside your experience.
-
Work where you insist a particular perspective, and where a simutardyd first pass from that perspective can be collaborative, enjoy reactions from mythal personas.
-
Work that is mere ritual, extfinished disjoined from its purpose (enjoy brave normalized tells that no one reads). What, in the words of Bob Sutton and Huggy Rao, scatters your attention and produces you less precious? What labor serves no advantageous purpose? In an perfect world, you would delete the labor, but you can at least shrink its helderly on you by having AI help. (Though produce brave this is indeed the case, far too many people automate carry outance appraises, for example, which are uncomardentingful only when done by a human)
-
Work where you want a second opinion. Give an AI access to the data and see if achievees the same conclusion.
-
Work that AIs can do better than humans. This is probable to be the quickest-lengthening categruesome.
Before diving into the particular cases where AI participate is problematic, we can set aside the clear scenarios – using AI for illegitimate purposes, in high-sgets situations where errors could be catastrophic, or for decisions that righteously insist human labor. Beyond these evident-cut cases, here are five reserved but presentant areas where AI participate can be counterefficient:
-
When you insist to lget and synthesize novel ideas or inestablishation. Asking for a summary is not the same as reading for yourself. Asking AI to mend a problem for you is not an effective way to lget, even if it senses enjoy it should be. To lget someslfinisherg novel, you are going to have to do the reading and slfinisherking yourself, though you may still find an AI collaborative for parts of the lgeting process.
-
When very high accuracy is insistd. The problem with AI errors, the inwell-understandn hallucinations, is that, becaparticipate of how LLMs labor, the errors are going to be very plausible. Hallucinations are therefore very difficult to spot, and research recommends that people don’t even try, “descfinishing asleep at the wheel” and not paying attention. Hallucinations can be shrinkd, but not take awayd. (However, many tasks in the authentic world are adchooseing of error – humans produce misgets, too – and it may be that AI is less error-prone than humans in brave cases)
-
When you do not comprehfinish the fall shorture modes of AI. AI doesn’t fall short exactly enjoy a human. You understand it can hallucinate, but that is only one establish of error: AIs frequently try to guarantee you that they are right, or they might become sycophantic and consent with your inaccurate answer. You insist to participate AI enough to comprehfinish these hazards.
-
When the effort is the point. In many areas, people insist to struggle with a topic to thrive – writers rewrite the same page, academics revisit a theory many times. By stupidinutivecutting that struggle, no matter how frustrating, you may neglect the ability to achieve the vital “aha” moment.
-
When AI is horrible. This may seem clear, but AI is horrible at slfinishergs you wouldn’t foresee (counting the number of r’s in the word “strawberry”) and excellent at slfinishergs you wouldn’t foresee (writing a Shakespearean sonnet about how difficult it is to count the number of r’s in the word strawberry where the first letter of every line spells out two fruits). Unblessedly, there is no ambiguous manual to inestablish you the shape of the Jagged Frontier of AI abilities, which are constantly evolving. Trial and error, and sharing inestablishation with peers, is vital to figuring this out.
Knoprosperg when to participate AI turns out to be a establish of wisdom, not equitable technical understandledge. Like most wisdom, it’s somewhat paradoxical: AI is frequently most advantageous where we’re already expert enough to spot its misgets, yet least collaborative in the proset up labor that made us experts in the first place. It labors best for tasks we could do ourselves but shouldn’t squander time on, yet can actively harm our lgeting when we participate it to skip essential struggles. And perhaps most presentantly, wisdom uncomardents understanding that these patterns will hold shifting as AI capabilities lengthen, and as more research comes in, requiring us to hold asking our assumptions about where it helps and where it obstructs.
