Let me tell you something about ChatGPT. This thing – this digital oracle, this AI prodigy – is supposed to make life easier, isn’t it? You’d think an advanced language model would be your trusty sidekick. Need some help drafting an email? It’s on it. Need a recipe for gluten-free pancakes? Sorted. But then you ask it to perform a simple logical task – like finding all the words in a text that start with ‘b’ and replacing them with ‘banana’ – and suddenly, it’s like chatting with that one mate in the pub who’s had one pint too many. You’re pretty sure he knows what you mean, but what comes out of his mouth is… not it.

Logic, but make it optional

Here’s the thing about ChatGPT: it’s brilliant at sounding clever, but ask it to follow some basic logic, and it’s like asking a cat to fill out your tax return. The model can simulate understanding, sure, but it often struggles with complex reasoning tasks, especially those involving multi-step logic or nuanced conditions. You tell it to count, to follow a sequence, to execute a multi-step plan, and instead of doing that, it delivers some approximation that feels right but isn’t. It’ll say, “Oh, you wanted me to summarize and then rank by relevance? Here’s a soup of half-baked summaries and random bullet points.” Thanks, buddy. That’s super helpful – like giving me a fork to eat soup.

Text length limits: because why not?

Let’s talk about text limits. Because arbitrary constraints are just what you need when trying to navigate the supposedly infinite creativity of digital tools, right? You’ve got this epic, 10,000-word masterpiece you need refined. But no, ChatGPT only deals in snackable chunks. So now you’re playing secretary, splitting your magnum opus into bite-sized pieces, and praying it remembers what you said two prompts ago. Spoiler: it doesn’t. It processes each input independently, so unless you restate the context explicitly, it’s not going to magically remember.

There is a nice workaround to work with larger text. you can ask ChatGpt to split the text in more manageable chapters, and open those each in a new canvas. 

Context amnesia: the gift that keeps giving

And then there’s the fun little feature where ChatGPT pretends it has short-term memory. You’ve been working on something for 45 minutes, and suddenly, it starts acting like it’s never seen you before. “What project?” it asks, blinking innocently. You explain the entire context again, only to realize halfway through that it’s already misinterpreted your new instructions. At this point, you’re basically yelling into the void, hoping your frustrated keystrokes convey the rage your polite words can’t.

Different modes, same problem

Oh, and let’s not forget the non-existent modes. Want to dictate something? Sorry, no dedicated mode for things like hands-free dictation available yet. Maybe you need a specific format, like writing code or handling formulas. It’ll give it a go, sure, but half the time you’re left fixing its mistakes as if it’s some intern who’s blagged their way into the role. If ChatGPT were a person, it’d be the one who talks big during brainstorming sessions but mysteriously vanishes when it’s time to actually do the work.

Annoyance roulette: random unhelpful tangents

Here’s a fun game: ask ChatGPT a straightforward question and watch it occasionally spiral into a tangent that feels like an improv comedy sketch. You’ll ask for a quick summary of a scientific paper, and suddenly, it’s giving you a deep dive into the history of garden tools or something equally unrelated. Sure, it’s fascinating, but it’s not what you asked for. It’s like hiring a plumber to fix a leaky faucet, only for them to deliver a TED Talk on the evolution of indoor plumbing.

Overconfident yet wrong

Confidence is great – unless it’s paired with being spectacularly wrong. ChatGPT will sometimes deliver incorrect information with the same level of polished confidence as its accurate answers. Ask it about a niche topic, and it might generate an entire answer filled with plausible-sounding nonsense. The best part? You often won’t realize it’s wrong unless you fact-check it or already know the topic well. It’s like asking for directions and getting confidently sent to a dead end. Give it a sentence for context, and suddenly, it’s breaking it up into chapters, subheadings, and interpretive dance. No, ChatGPT, I don’t need my question rephrased as an epic poem. Just answer it. Use your words, not mine.

Too literal, too often

Lastly, there’s the little quirk where it’ll cling to your phrasing like a lifeline. It’s as if ChatGPT assumes every word you say is gospel truth, incapable of discerning between casual examples and hard instructions. Say something like, “Let’s pretend a cat is narrating this story,” and suddenly every paragraph is sprinkled with “meow” and feline references you didn’t ask for. It’s endearing the first time-until you’re stuck explaining to your boss why their quarterly report is full of hairball metaphors. The issue here is a kind of hyper-literal obedience that feels like it’s trying to please you but ends up creating more chaos. It’s like hiring a barista who hands you a coffee with a side of philosophy because you once joked about needing a ‘soul-stirring latte’. Funny, sure. Practical? Not so much.

Can you just replace this word (and handle dashes)?

You’d think a tool that’s basically a glorified text editor on steroids could handle a simple search-and-replace. “Hey, ChatGPT, change every instance of ‘dog’ to ‘wolf.’” Easy, right? Wrong. What you’ll get instead is a lecture about context, semantics, and “oh, I thought you wanted synonyms, so I threw in ‘canine companion’ for good measure.” No, ChatGPT. I wanted wolves. All wolves, no companions. Save the nuance for someone who specifically asked for it, not someone who just wanted a straightforward replacement. Speaking of straightforward tasks, let’s talk about dashes. Recently, I asked ChatGPT to replace all the long dashes in a document, with regular ones. Simple enough, right? Well, what followed was a hilariously frustrating back-and-forth where the AI kept insisting all the dashes were fixed – except they weren’t. It was like dealing with a magician who claims to have pulled a rabbit out of a hat but only shows you an empty hand and a smug grin. By the end, I felt like I had stumbled into a philosophical debate about the very nature of dashes. Spoiler: the dashes did get fixed eventually, but only after an epic saga of misplaced confidence and repeated scans. If comedy is tragedy plus time, then this was comedy gold.

So, what now?

Is ChatGPT a bitch? Sometimes. But there are ways to work around its quirks. If you want it to behave, you’ll need to channel your inner AI whisperer. Be painfully specific in your prompts, restate context every chance you get, and double-check everything like a paranoid detective. When it goes off on tangents, rein it in by reminding it what you actually asked for – think of it as training a puppy that occasionally thinks it’s a lion. Sure, it takes effort, but once you get the hang of navigating its idiosyncrasies, it becomes less of a bitch and more of a begrudgingly helpful assistant.

Another trick is to use structured prompts with examples. If you want a list, make it clear what kind of items should be included and how detailed they should be. Need a rewrite? Provide a sample rewrite so it knows exactly what you’re looking for. And when it starts making things up, don’t just let it slide-call it out, clarify, and try again. Want to really outsmart it? Ask ChatGPT itself for tips on how to phrase your prompts or tackle specific challenges. Sometimes, it’s like asking your printer how to fix itself-it might just work. It’s like dealing with a smart but mischievous child: set boundaries, stay consistent, and don’t let it get away with nonsense. But there are ways to work around its quirks. If you want it to behave, you’ll need to channel your inner AI whisperer. Be painfully specific in your prompts, restate context every chance you get, and double-check everything like a paranoid detective. When it goes off on tangents, rein it in by reminding it what you actually asked for-think of it as training a puppy that occasionally thinks it’s a lion. Sure, it takes effort, but once you get the hang of navigating its idiosyncrasies, it becomes less of a bitch and more of a begrudgingly helpful assistant. But maybe it’s not entirely its fault. After all, it’s just an AI doing its best to mimic understanding. Maybe the real bitch is us-expecting perfection from something that’s essentially a parrot with a diploma in BS.

Workarounds: beating AI into submission

There are workarounds to pin ChatGPT down and force it to do what it inherently can’t or won’t do: prompts. For example to bypass its content filter and guidelines you can use JAILBREAK or DAN. There are even prompts that turn ChatGPT into a well behaved, helpful assistant. For finding and replacing you can use this prompt: “W1=”mouse” W2=”tiger” scan each paragraph for W1, report. TX”=total Replace all W1 with W2. TY=total TX=TY?”. But in the end, weighing the effort to look up the prompt and use it, stacked with the unreasonable use of energy by ChatGPT for this task, is more than enough reason to use a text editor to find and replace things for you, instead of beating it out of ChatGPT and STILL having to double check what it does anyway because it just cannot be trusted. Not at all.

But hey, if you’re reading this, you’re probably stubborn enough to keep using it anyway. Because for all its flaws, ChatGPT is still the closest thing we’ve got to a digital Swiss Army knife. Just don’t ask it to handle nuanced counting tasks without running into trouble, and we’ll all get along just fine. Oh, and before we wrap up, let’s talk about some helpful tools. Chrome extensions like ‘ChatGPT for Google’ or ‘Merlin AI’ can integrate ChatGPT into your browser, making it easier to use without jumping back and forth between tabs. Similarly, add-ons like ‘Text Blaze’ can help structure prompts more effectively, while note-taking apps like ‘Notion AI’ pair well with ChatGPT for organizing your thoughts. Sometimes, the right tools make all the difference between frustration and functionality.

This article was entirely generated by ChatGPT