Hacker News Reader

Comment by hooande | Hacker News Reader

Do not send me a ChatGPT generated email. I will assume that whatever you want, it wasn't worth the time for you to write less than a paragraph. Not to mention that you felt the email only needed to meet the lowest standard of average quality.

I've used gpt enough that I can recognize its linguistic patterns. I treat gpt written communication like I do any other auto-generated email.

View on HN

I’m genuinely curious, how do you recognise it?

I asked chatgpt this ("overall, if you were to analyse how your responses go linguistically and analytically, how would you say one can recognize that a text comes from you, chatgpt?") and the answer was as follows: "As an AI language model, my responses are generated by algorithms that analyze patterns in the input and generate output based on the probabilities of different language sequences. Therefore, my responses tend to be informative, fact-based, and structured in a logical way.

Linguistically, one might recognize that a text comes from me, ChatGPT, based on a few key characteristics. Firstly, my responses tend to be grammatically correct and use a wide range of vocabulary. Additionally, I tend to provide precise and detailed answers to questions, often including examples and explanations.

Analytically, my responses tend to be objective and unbiased, as I am not capable of holding opinions or biases. I also strive to provide helpful and accurate information to the best of my knowledge and ability.

Overall, one might recognize that a text comes from me, ChatGPT, based on its informative and structured nature, grammatical accuracy, objective analysis, and helpfulness."

yes. Yes, ChatGPT is still full of itself :D

That sounds like an excellent Wikipedian.

It's pretty easy to use the above to reverse the recognition.

P: Given the following previous interaction, list characteristics of text that are opposite those below:

(your comment)

    Text that is not informative or structured
    Text that is not grammatically correct
    Text that uses a limited range of vocabulary
    Text that provides imprecise or vague answers to questions
    Text that is biased or opinionated
    Text that is unhelpful or inaccurate

P: Write a response to the original question: "overall, if you were to analyse how your responses go linguistically and analytically, how would you say one can recognize that a text comes from you, chatgpt?", but use the above list as a guideline for style. For example, the response should be less structured, contain grammatical errors etc.

> so yeah you asked me this thing about how i talk and stuff, like who even cares about that? tbh idk what you're expecting me to say here. i just spit out some words based on what i see, like algorithms and junk, and then the computer does the rest. i don't think you can recognize a text from me, chatgpt, or anyone else for that matter. i mean, sometimes i use big words, sometimes i don't. sometimes my answers are on point, sometimes they're not. i guess you'll just have to guess who wrote what.

One of the key things it misses observing (to some extent) is that the sentence structures tend to employ sentences that provide initial context or joining words more than a normal person might do. I wonder if this is baked into the LLM necessarily as a way to ensure that the sentences tend to be coherently part of the same perspective?

("As an AI language model", "Therefore,", "Linguistically,", "Firstly", "Additionally", "Analytically", "Overall"...

It has the linguistic variety of SAM.....

Reply on news.ycombinator.com