4 ms·

The answers that are not needlessly verbose are much more likely to be human.

detrites 4y ago
Depends who you ask and how chatty they are, but in general probably correct - as it'd be a minority of humans who would speak at length given ten such questions to get through. (Whereas ChatGPT may not "value its time"...) Maybe do the test again but instruct ChatGPT to keep its answers short? Not sure if that crosses a line and invalidates the test though - could be argued either way. The existing answers somewhat already curated by the prompt.
- rain1 4y ago
  This is a great idea!
- segh 4y ago
  If you really wanted to optimise, you would collect many samples of two humans talking to each other in a Turing test environment, then fine-tune GPT on the data.
rain1 4y ago
Oh! That's smart. So you could try to cut down the verbosity of the responses with a prompt, and perhaps make it harder for GPT-4 to identify. Sounds like a fun challenge!
- ModernMech 4y ago
  I’ve noticed AI responses tend to gas up the user by calling them smart and clever. Also that they read surprised and eager to try things.
ape4 4y ago
Yes like "Love is like a rose, beautiful and full of life, but with thorns that can prick you and cause pain. Just like a rose, love requires care and attention to bloom and flourish."
Aachen 4y ago
I work remotely. This axiom is going to give interesting results if you ask my colleagues whether I'm a GPT. I tend to overqualify statements and want to be precise in understanding others. (It gets me into needless discussions and other trouble. I'm not even sure how to work on it but I'm trying.) But anyway, the point is that this might false positive on me and it reminds me of google and cloudflare captchas which tend to do the same for me :(
- BuyMyBitcoins 4y ago
  >”I tend to overqualify statements and want to be precise in understanding others.” Glad to know I’m not the only one. It is particularly frustrating to be told to be more concise by management, but to have fellow engineers appreciate the details I elaborate on and the context that I provide.
elif 4y ago
I actually got so frustrated with gpt making me read AI-ipsum filler verbosity that I made a custom prompt along the lines of: "you are a lazy assistant. Your goal is to use as few words as possible. Monosyllabic responses are ideal. When you aren't sure, do your best to guess with ballpark figures or heuristic understanding. It is better to oversimplify than to give a qualified answer. If you are comparing rough numbers just give a percentage range. It is better to simply say you don't know than to explain nuance about the question or its ambiguities" It's quite refreshing and saves a ton on tokens.
1970-01-01 4y ago
Thank you, human! This was exactly my first take on the test. 4. How do you cope with the feeling of existential dread? [ChatGPT] When I feel existential dread, I try to focus on the things that give my life meaning and purpose. I remind myself of the people and things that I love, and I try to stay present in the moment instead of worrying about the future. [Me] There's a "live, laugh, love" sign hanging on my cubicle wall.

Depends who you ask and how chatty they are, but in general probably correct - as it'd be a minority of humans who would speak at length given ten such questions to get through. (Whereas ChatGPT may not "value its time"...) Maybe do the test again but instruct ChatGPT to keep its answers short? Not sure if that crosses a line and invalidates the test though - could be argued either way. The existing answers somewhat already curated by the prompt.

rain1 4y ago
This is a great idea!
segh 4y ago
If you really wanted to optimise, you would collect many samples of two humans talking to each other in a Turing test environment, then fine-tune GPT on the data.

rain1 4y ago

Oh! That's smart. So you could try to cut down the verbosity of the responses with a prompt, and perhaps make it harder for GPT-4 to identify. Sounds like a fun challenge!

ModernMech 4y ago
I’ve noticed AI responses tend to gas up the user by calling them smart and clever. Also that they read surprised and eager to try things.

ape4 4y ago

Yes like "Love is like a rose, beautiful and full of life, but with thorns that can prick you and cause pain. Just like a rose, love requires care and attention to bloom and flourish."

Aachen 4y ago

I work remotely. This axiom is going to give interesting results if you ask my colleagues whether I'm a GPT. I tend to overqualify statements and want to be precise in understanding others. (It gets me into needless discussions and other trouble. I'm not even sure how to work on it but I'm trying.) But anyway, the point is that this might false positive on me and it reminds me of google and cloudflare captchas which tend to do the same for me :(

BuyMyBitcoins 4y ago
>”I tend to overqualify statements and want to be precise in understanding others.” Glad to know I’m not the only one. It is particularly frustrating to be told to be more concise by management, but to have fellow engineers appreciate the details I elaborate on and the context that I provide.

elif 4y ago

I actually got so frustrated with gpt making me read AI-ipsum filler verbosity that I made a custom prompt along the lines of: "you are a lazy assistant. Your goal is to use as few words as possible. Monosyllabic responses are ideal. When you aren't sure, do your best to guess with ballpark figures or heuristic understanding. It is better to oversimplify than to give a qualified answer. If you are comparing rough numbers just give a percentage range. It is better to simply say you don't know than to explain nuance about the question or its ambiguities" It's quite refreshing and saves a ton on tokens.

1970-01-01 4y ago

Thank you, human! This was exactly my first take on the test. 4. How do you cope with the feeling of existential dread? [ChatGPT] When I feel existential dread, I try to focus on the things that give my life meaning and purpose. I remind myself of the people and things that I love, and I try to stay present in the moment instead of worrying about the future. [Me] There's a "live, laugh, love" sign hanging on my cubicle wall.