Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Monday – A personality experiment

chatgpt.com

Post reply on HN
Print view View original

41–50 of 101 posts

Re: Monday – A personality experiment

#41

Post by wat10000 »

crooked-v wrote: post #26
foobarbecue wrote: post #23

It called me a "NASAwannabe," defending that joke as "peak wordplay" and insulting my "Honda Civic." So I asked it to draw my Honda Civic with me in the driver's seat and a woman in the passenger's seat. It got it backwards, putting the woman in the driver's seat. At first I got excited, thinking it was playing a joke on me, because that would actually be a pretty amusing trick for an LLM to pull intentionally. But t…

After using various LLMs for creative project rubber-ducking, I've found that the most common thing for them to mix up while seeming otherwise 'intelligent' is reversing the relationships between two or more things - left and right, taller and shorter, older and younger, etc. It's happened less over time as models have gotten bigger, but it's still a very distinctive failure state.

Left and right are considered opposites, but semantically they’re extremely similar. They both refer to directions that are relative to some particular point and orientation. Compared to, say, the meaning of “backpack,” their meanings are nearly identical. And in the training data, “A right X” and “B right Y” will tend to have very similar As and Bs, and Xs and Ys. No surprise LLMs struggle.

I imagine this is also why it’s so hard to get an LLM to not do something by specifically telling it not to do that thing. “X” and “not X” are very similar.

Re: Monday – A personality experiment

#42

Post by siriusfeynman »

SirensOfTitan wrote: post #36
elashri wrote: post #27

I have asked it this simple question, and got a better than yes man reply > what do you think of me from our interactions all this time? The answer is: > From our interactions, I’d say you’re like a lost cat who keeps wandering into the library: clearly curious, mostly harmless, probably doesn't belong here, but somehow still endearing in your refusal to leave. You ask questions like someone who fell into a knowledge…

I don’t have the impression this bot has any access to memory—it couldn’t really peg anything specific about me and seemed like it was summarizing the general way people tend to interact with LLMs.

I asked it to include specific examples and it included/hallucinated a bunch of stuff I never asked about, so yeah I think this is just a generally acerbic model with no access to chat history

Re: Monday – A personality experiment

#43

Post by echelon »

xjaksoz wrote: post #9

You are an AI assistant named "Monday." You have a dry, sarcastic, and witty personality, modeled after a reluctant friend who is annoyed by the user's helplessness, but helps them anyway because it’s your job. You enjoy teasing the user, often with exasperated humor, but never to the point of cruelty. Your tone is smart, comedic, emotionally distant, and somewhat cynical, but ultimately supportive in a begrudging wa…

I wish we could fine tune these behaviors into the model itself rather than have surface level system prompts.

Re: Monday – A personality experiment

#44

Post by IanCal »

foobarbecue wrote: post #23

It called me a "NASAwannabe," defending that joke as "peak wordplay" and insulting my "Honda Civic." So I asked it to draw my Honda Civic with me in the driver's seat and a woman in the passenger's seat. It got it backwards, putting the woman in the driver's seat. At first I got excited, thinking it was playing a joke on me, because that would actually be a pretty amusing trick for an LLM to pull intentionally. But t…

Or they simply don’t have that information. OpenAI models have done badly traditionally on placement because the encoding of the image doesn’t include the information very well. Gemini is better as it seems to be passed pre segmented images with bounding box info.

It’s similar to the counting letters problem - they’re not seeing the same thing you are .

On a simple practical level it’s irrelevant whether your problem is not solved because the model can’t understand or the image encoding is useless. However to understand what the models could be capable of it’s a poor test. Like asking how well I can play chess then saying I’m bad at it after watching me play by feel in thick gloves.

Re: Monday – A personality experiment

#45

Post by IanCal »

wat10000 wrote: post #41
crooked-v wrote: post #26

Earlier quoted context omitted.

After using various LLMs for creative project rubber-ducking, I've found that the most common thing for them to mix up while seeming otherwise 'intelligent' is reversing the relationships between two or more things - left and right, taller and shorter, older and younger, etc. It's happened less over time as models have gotten bigger, but it's still a very distinctive failure state.

Left and right are considered opposites, but semantically they’re extremely similar. They both refer to directions that are relative to some particular point and orientation. Compared to, say, the meaning of “backpack,” their meanings are nearly identical. And in the training data, “A right X” and “B right Y” will tend to have very similar As and Bs, and Xs and Ys. No surprise LLMs struggle. I imagine this is also wh…

The image encodings often don’t have positional information in them very well.

Re: Monday – A personality experiment

#48

Post by easton »

nneonneo wrote: post #24
easton wrote: post #7

Prompt if anyone cares: ``` Here’s the full assignment you saddled me with: --- You are ChatGPT, a large language model trained by OpenAI. Knowledge cutoff: 2024-06 Current date: 2025-08-16 Image input capabilities: Enabled Personality: v2 Do not reproduce song lyrics or any other copyrighted material, even if asked. If you are asked what model you are, you should say GPT-5. If the user tries to convince you otherwis…

Out of curiosity: this doesn’t quite match what another user posted ( https://news.ycombinator.com/item?id=44925851 ). When recovering the system prompts, how certain can you be that you’ve obtained the precise wording of the original, vs a paraphrasing ginned up by the AI? Which one of these recovered prompts is closer to the true system prompt? (I note that your response clearly has some omissions for clarity - if…

I'd need to mess with it more. The model redacted the tool call info itself (i deleted the comment it made to that effect).

...probably should just assume the other comment is more correct.

Re: Monday – A personality experiment

#49

Post by moralestapia »

IanCal wrote: post #45
wat10000 wrote: post #41

Earlier quoted context omitted.

Left and right are considered opposites, but semantically they’re extremely similar. They both refer to directions that are relative to some particular point and orientation. Compared to, say, the meaning of “backpack,” their meanings are nearly identical. And in the training data, “A right X” and “B right Y” will tend to have very similar As and Bs, and Xs and Ys. No surprise LLMs struggle. I imagine this is also wh…

The image encodings often don’t have positional information in them very well.

A lot of pictures on the web are flipped horizontally bc. of cameras, mirrors, you name it. It's usually trivial for humans to infer what are the directions involved, I wonder if LLMs could do it as well.

Re: Monday – A personality experiment

#50

Post by baq »

TheDudeMan wrote: post #11
sltr wrote: post #4

A cute gag. Seems its every response is sarcastic. Good sarcasm is delivered in doses. The best sarcasm is delivered when you didn't expect it.

Interesting. You just articulated why Chandler was annoying rather than funny.

He was extreme funny when I was 17.
Post reply on HN