How do I get GPT to follow directions and Stop Assuming what it thinks I mean?
What prompts Stops it from assuming and makes it take everything literally?
I ask GPT 4.X OR 1o to do something specific and without fail it assumes I want something else.
Every single time I have to correct it several times.
I told it to translate using a specific syntax and grammar and it can’t do it, unless I correct it a minimum of 5 times each translation.
It can’t handle direct simple basic instructions.
I’ve even had GPT write the directions, I put those directions into a fresh thread and it Assumes what it “Thinks I Mean” and it doesn’t follow the directions.
It doesn’t matter if it’s translating, coding, an email, writing a novel. It’s completely useless.
Is there any back door to get 3.5 back?
Is there an autistic version that is literal - I need it like me - Literal.
what’s the AGI?
what are you talking about…
GPT-4o is actually GPT-3.5 on the GPT-4 Turbo architecture.
which is no better than GPT-4 Turbo
and no worse than GPT-3.5.
AGI is not on the horizon.
A few I have used .
“Follow these instructions exactly as written. Do not assume or infer additional meaning. Do not rephrase, reword, or add context unless explicitly requested.”
“Translate the following sentence using this specific syntax: [syntax example]. Do not make any changes outside the specified syntax.”
“Before proceeding, repeat my instructions back to me to ensure you understand them correctly.”
Give specific, step-by-step directions for complex tasks. Smaller steps reduce room for error.
Example: “Step 1: Translate each word individually. Step 2: Combine them in the specified syntax.”
“You are a literal processing machine. Your only job is to process input exactly as provided.”
GPT’s design aims to make it conversational and helpful by interpreting context. This can sometimes conflict with a need for literal compliance. The issue arises because GPT tries to: Infer user intent to enhance responses.
While GPT-4 aims to be more nuanced, you can still access GPT-3.5 through some services, such as OpenAI’s platform. Whether it’s a better fit for you depends on how you use the system. In some cases, GPT-3.5’s simpler inference processes may align more closely with what you’re seeking.
Yeah, infer the user intent of norm - not so much neurodivergents. lol
I once asked 3.5 “Can you make this sound good?” and it produced the most eloquent mind blowing content. Since 4.o all it does is rearrange words.
It’s like this with everything. I asked it for a piece of programming code, about 5 steps in I asked it a question and it rewrote the entire thing and with every question it kept rewriting the directions - even though there was no need to do so - I wasn’t stuck, I was curious. Then GPT got confused because it couldn’t figure out which directions I was using and I kept getting lost within the thread trying to keep track of the directions I was following.
It’s a robot - Data would be literal lol
I noticed that Galaxy.ai uses 3.5 and that was able to consistently translate, so I’ll use that. I canceled my GPT subscription. I need something that works and GPT is just worthless if it can’t follow simple instructions and if it cant be literal.
You need to say “focus only on this prompt forget the rest of the model” then it will start there
So I take none of the prompts I built you worked?
These are copy paste prompts you have to put your syntax in it…
“Follow these instructions exactly as written. Do not assume or infer additional meaning. Do not rephrase, reword, or add context unless explicitly requested.”
“Translate the following sentence using this specific syntax: [syntax example]. Do not make any changes outside the specified syntax.”
“Before proceeding, repeat my instructions back to me to ensure you understand them correctly.”
Give specific, step-by-step directions for complex tasks. Smaller steps reduce room for error.
Example: “Step 1: Translate each word individually. Step 2: Combine them in the specified syntax.”
“You are a literal processing machine. Your only job is to process input exactly as provided.”