I commemorate whenChatGPTfirst appear , and the first thing everyone started saying was “ Writers are done for . ” People part speculating about newsworthiness internet site , blogs , and middling much all write cyberspace content becoming AI - generated — and while those predictions seemed utmost to me , I was also pretty impressed by the school text GPT could produce .
Naturally , I had to strain out the fancy young tool for myself but I quickly discovered that the results were n’t quite as telling as they seemed . tight onward more than two years , and as far as my experience and my use cases go , nothing has changed : whenever I use ChatGPT to help with my writing , all it does is slow me down and leave me frustrated .
Why ChatGPT just doesn’t work for me
I ’m meritless , I simply can not be misanthropic about a engineering science that can accomplish this.pic.twitter.com/yjlY72eZ0 m
When you take for something like that , part of what make the result bang-up is that it ’s unexpected . You ’ve basically asked it to storm you , and in most cases , it will succeed .
When you ’re trying to use ChatGPT for boring old writing workplace , on the other hand , it ’s a whole dissimilar tale . First of all , for pieces like this , it ’s useless . Everything I ’m writing now is about my own experience and opinion — AI ca n’t publish those for me . Some hoi polloi might indicate that it could help you design , insight , or refine arguments — but I ’ve never wanted help with that kind of thing anyway .
Willow Roberts / Digital Trends
The kind of study I actually tried to use ChatGPT for was troupe blog . You know the type — explainers , how - tos , and recommendation mail brood matter come to to the fellowship and its product ( with some elusive ego - promotional material thrown in as well ) . When you write this kind of matter , you ’re often given a lot of prerequisite : a stylus usher for language and grammar , keywords to slip in , sources to include , sources to avoid , and a content outline with the bearing , anatomical structure , and cardinal points all pre - decided .
If I wanted to get some operational copy , I could n’t just flow GPT a one - line command prompt and allow it course baseless . So I strain a social system that depend like this :
I would never necessitate for too much at once and since I did n’t entrust it to add stats and source , I ’d leave all that out with the intention of doing it myself after .
Willow Roberts / Digital Trends
But as much as I seek to split up the work into diminished chunks and give profusely clear instructions , I would always run into the same problems :
get me show you what I mean .
1. It’s terrible at listening to instructions
When you ’re trying to get very specific output from ChatGPT , you have to give specific education . Unfortunately , it feel like the more thing you take for , the more likely GPT is to snub some of them . My prompts would have headings , content hummer points , intelligence counts , initialise instructions , and the chatbot had to remember the trend instructions from the startle of the school term too . I tried band of dissimilar glide slope to simplify things but it always mat like ChatGPT just could n’t cover this many instructions .
Two specific things I would have problem with frequently were word counts and heater points . The LLM rarely pass on me the number of intelligence I asked for ( unremarkably giving me something way under rather than way over ) , and it never listened to me when I said I did or did n’t want any bullet points .
Sometimes this is fixable — say I want 200 words with bullet train point but I make 300 discussion without bullet points . It ’s somewhat quick to cut a few countersign and thrust some bullet point in there . regrettably , it always felt like I got the harder - to - fix mistakes .
When you inquire for 500 intelligence with no bullet point and get 200 words with slug point — you basically have to do most the work yourself .
2. It has a bad habit of taking things too far
When you tell ChatGPT you require something specific , such as friendly language or second individual point of view , it tends to latch onto these conception and go sick with them .
Friendly language turns into full - on text edition chat language with emoji , and 2nd - somebody linear perspective somehow wrench into excessive questions and point of reference to the lector so it can use the pronoun “ you ” as often as possible .
You ’re also a snatch stuck if you want a lot of something like illustration or quotes . If you just tell GPT you want “ lots , ” “ plenty , ” or “ many ” of something , it will probably give you dual or triple the amount you need . If you give it a specific number , it ’s potential to ignore it completely and give you something random . In other language , you’re able to scarce control the output .
3. Incorrect or irrelevant information sneaks into the copy quite often
We all know that AI “ hallucinates ” and gets its facts wrong at random times — it happens enough that you have to see every single affair it says , which takes a long meter and affects how much sentence you could save by using it .
To battle this , I essentially never expect for facts or figures . I did try a few times right at the beginning but the fact - curb process really is a slog , and I quickly block up get to .
The trouble is , GPT will throw random inaccuracy into its answer whether you take for fact and figures or not . This think you ca n’t just check out how well the written matter reads or whether you ’re happy with the gunpoint it ’s trying to make — you have to check and consider the rigor of everything it says . It ’s such a drag .
And since I ’m blab out about mistakes and hallucination here , I might as well mention the “ worst - case scenario ” too . Sometimes the LLMjust goes off the rails , and while this does n’t happen all the time , when it does — you ’ve obtain to throw that session in the bin and expend time pasting your prompting into a newfangled Old World chat and starting all over . I ’ve never really consider ChatGPT go crazy in a particularly funny way personally , but my friend did get this stone once :
4. Its “writing style” gets repetitive and cliche very quickly
At first , it seems insane just how “ human ” ChatGPT can voice . But as you utilize it more and more , you realise it does n’t just sound human , it sounds like “ the mean human . ”
OpenAI ’s data sets include most of the internet — millions of cyberspace articles , Reddit togs , and personal blogs . And I ’m sad to say , a peck of this content is gross garbage . But because ChatGPT is trained on it , it blame up all of the most common bad riding habit .
So when you generate a tidy sum of text with ChatGPT , you ’ll set out notice that sealed judgment of conviction structures and musical phrase get repeated a lot . The two worst perpetrator for me were these two sentence structure :
ChatGPT like these two sentence structures so much that I was pretty much guaranteed to get three or four of them in every single session . Those two examples even came from the same paragraph of one of its responses . And that “ In the world of … ” phrase in the first illustration is another way it really fuck to pop out a judgment of conviction . All of it is boring , banality , and preposterously overused ( which is , of line , the very cause ChatGPT finish up give them so much ) . I even seek expressly banning sure phrase and sentence type , starting each prompting with this little lean :
multitude say it ’s almost unacceptable to say AI - generated text from man - write text nowadays — but I kind of disagree . If you ’ve attempt to use theseAI toolsyourself and experienced all of the problem and bad habits , you get to know a set of tell - story signs . When people mix AI content with human - written hooey and heavy edit the absolute majority of ChatGPT ’s output , you may hide it completely . But content that ’s just come straight from the language manakin and published practically as is — you’re able to severalise . you could tell quite easily .
5. It’s near impossible to get the LLM to revise or correct its output successfully
The material muckle breaker with all of this is when a problem occurs , ChatGPT can rarely fix it alone . Whenever thing get amiss , I would taste once or twice to explicate the problem and ask for revisions , but it just did n’t work .
If I inquire for the right Bible count , most of the time I would just get the same word count again . If I asked it to get free of the bullet points , it would say “ Of course ! ” and then give me more slug points . If I call for it to adjust the tone or the style , it would clamber to apply the change across the table , and I ’d end up with a weird intermixture of both .
possibly if you just kept reprompting and regenerating ChatGPT ’s responses , you would get moderately close to what you wanted finally . The problem for me is that I ’m a writer — and using ChatGPT forces me to fill the editor role instead .
This probably is n’t a problem for everyone — peck of author also do a bit of editing as part of their work , but it really bothers me . I hate editing other people ’s work and I detest editing GPT ’s output .
As for how often things went wrong — when every academic session is multiple hours long with 30 + command prompt , you nearly always strike a snag somewhere .
The result: I gave up
I did essay for a skilful few months to learn how to get some use out of ChatGPT back in 2023 and I conk out back to it after major updates over the next two days — but my experience never changed . I tried other LLMs too , but even the new “ reasoning ” framework that blurt out out interior soliloquy before answering still have the same serviceableness issues .
Current LLM modelsjust do n’t speed up the writing process for me — all they do is force me to spend time doing what I hate instead of just couch time into what I enjoy . If you detest writing and never want to do it , ChatGPT can most surely help you out . But if writing is a hobby for you or what you ’ve choose to do as a living , this affair will likely drive you crazy .
In the end , I never published anything one could call “ AI - generated . ” Every time I essay to , as they say , “ integrate it into my workflow , ” I would terminate up bash my head against the paries , wasting a heap of time , and then closing the thing down when I clear I had accomplish nothing and my deadline was just around the recession .
