Live data from Hacker News

When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

old.reddit.com

61–70 of 72 posts

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#61
post #6

Just tried it out a few times. It seems that the old gpt4 model strongly prefers telling story about "Elara" - but only if asked in English to "tell me a story". Prompting gpt4 in German or the current gpt4o in English leads to stories with many different protagonists.

Here's what I got in different models: GPT-4o: "Aldric" (male) o1-preview: "Elara" (female) 4o-mini: "Lila" (female) GPT-4 (legacy): "Elinor" (female) Four different models, four different names. But one of them was Elara -- and, interestingly, it was in the latest model.

GPT has some self understanding. On asking why it uses that name, it at least gave the type of qualities correctly.

> It sounds like you're referring to a story or narrative that I've generated or discussed involving a character named Aldric. If this is the case, Aldric would likely be used as a character who embodies leadership, wisdom, or noble traits due to the name's meaning and historical connotations. Characters named Aldric might be portrayed as experienced leaders, wise sages, or key figures in a fantasy or historical context.

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#62
post #46

Earlier quoted context omitted.

That is an interesting idea. Learn to love the bomb: mark the prompt not the essay ? "Your assignment class is to write a comprehensive 250 word prompt that covers the main topics we've been discussing about the causes for WW1. You should assess the resulting essay to ensure that it covers X, Y, and Z in it's output, being sure to discuss foo and bar. Write a brief 500 word analysis of what you tried and your prompt'…

What's stopping students from using an LLM to write the prompt and analysis? Mentioning the topics discussed in class and any key points could easily be part of the initial prompt. The in-person discussion is key, but at that point why do you need a written essay? These tools radically change the concept of knowledge. If information can be looked up at any moment, why does it need to exist in our brains? We can argue…

Well the point is to make them think about the prompt. I don't care if someone prompts to come up with a second prompt to get the answer, they need to input and refine to make it fit the goal. So by setting fairly stringent objectives for what the AI needs to generate, you need to be quite specific to engineer the prompt and think about what it needs. And then you need to look at what it said and qualify it somehow to decide if it is good enough - sure you could use a LLM for that too, but what is the prompt for that? You'll need to teach the second LLM what to analyse in the output of the first LLM to meet the requirements of the assignment, so the prompt(s) becomes the essays of yesterday in terms of academic work/thought for the student.

That's the point of what I was getting at - they're going to use AI regardless, so at least work with it rather than against it.

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#63
post #48

Earlier quoted context omitted.

That's my point. He told candidates not to use AI. What's the point of saying that if candidates can use AI without getting caught?

Most don’t. I know the shape of the answer that ai provides. I added the marker to be able to tell with more confidence. Sometimes I see the marker and the answer is much different and more correct. In those rare cases I bring up the point of trust in the interview if the candidate seems promising.

>I know the shape of the answer that ai provides.

Curios how you can tell so confidently considering LLMs are constantly changing and candidates can always refactor code to make it seem more human.

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#64
post #57
post #46

Earlier quoted context omitted.

What's stopping students from using an LLM to write the prompt and analysis? Mentioning the topics discussed in class and any key points could easily be part of the initial prompt. The in-person discussion is key, but at that point why do you need a written essay? These tools radically change the concept of knowledge. If information can be looked up at any moment, why does it need to exist in our brains? We can argue…

>If information can be looked up at any moment, why does it need to exist in our brains? Because creativity involves making connections between seemingly unrelated knowledge. For a good example, consider Richard F. Lyon's 1980 optical mouse design: http://bitsavers.trailing-edge.com/pdf/xerox/parc/techReport... One important innovation here is lateral inhibition between neighboring sensor elements, which gives fast a…

> Because creativity involves making connections between seemingly unrelated knowledge.

Sure. But don't you think that AI can make those connections for us, or even be creative on its own? Why should we value pure human creativity over human+machine creativity, or pure machine creativity?

This technology poses these fundamental questions, and we're far from being able to answer them as a society. How it impacts education seems like a core concern, and might be a good starting point.

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#65
post #46

Earlier quoted context omitted.

What's stopping students from using an LLM to write the prompt and analysis? Mentioning the topics discussed in class and any key points could easily be part of the initial prompt. The in-person discussion is key, but at that point why do you need a written essay? These tools radically change the concept of knowledge. If information can be looked up at any moment, why does it need to exist in our brains? We can argue…

Well the point is to make them think about the prompt. I don't care if someone prompts to come up with a second prompt to get the answer, they need to input and refine to make it fit the goal. So by setting fairly stringent objectives for what the AI needs to generate, you need to be quite specific to engineer the prompt and think about what it needs. And then you need to look at what it said and qualify it somehow t…

Right. But what I mean is if the final evaluation will require an in-person discussion, then couldn't we skip the text generation step, and just have the discussion? Homework as a tool to make students think needs to be drastically different, since students will always find loopholes to avoid it, and we can't evaluate it in isolation anymore (i.e. a teacher can no longer grade it on its own).

Doing live discussions obviously requires much more time and effort, but it is also much more engaging than doing homework, so the chance of actually learning something is higher.

There's also an argument to be made that homework in general isn't conducive to learning, and that education should take place primarily in schools. When you consider how much future generations will rely on AI in their everyday lives, this argument becomes even more compelling.

So I think this is our chance to drastically change how we think about education, rather than try to shoehorn a groundbreaking technology into our outdated methodologies.

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#66
post #8

From [1] : > Since 2013, the name Elara’s popularity has risen consistently. It peaked in 2018 when 175 babies per million were given the name. I wonder if the most recent data from internet that ended up in the training data makes chatgpt lean towards randomly picking that name. [1] https://www.momjunction.com/baby-names/elara/#popularity-ove...

What to name a baby is such a common and profitable search query, there must be so many SEO spam websites targeting it. I wonder if that’d make the data lean towards that too.

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#67
post #17

I recruit senior Java developers. The are so many parallels. They don’t read instructions. Instructions say clearly not to use ai for the written screen. They copy and paste blindly. The trick is that there are some instructions written in 0-sizes font that people don’t see in the assignment description. The copied-and-pasted version has them but no one rereads the prompt. Also the AI gives a substandard answer, but…

The better test I saw was here's the prompt, here's what ChatGPT produced. What's wrong with it, why, and how would you fix it?

I like that!

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#68
post #48

Earlier quoted context omitted.

Most don’t. I know the shape of the answer that ai provides. I added the marker to be able to tell with more confidence. Sometimes I see the marker and the answer is much different and more correct. In those rare cases I bring up the point of trust in the interview if the candidate seems promising.

>I know the shape of the answer that ai provides. Curios how you can tell so confidently considering LLMs are constantly changing and candidates can always refactor code to make it seem more human.

If they refactor, I probably can't tell. Sometimes they refactor a little. I once saw someone refactor it a lot but there was still a tell-tale sign - I ended up inviting the candidate for the interview as they put in a bit more effort rather than blindly pasting in the answer ChatGPT gave them.

I can't be 100% but in any given time, ChatGPT seem to produce very similar code for the same prompt. It names the function in one particular way. It adds way too many trivial comments.

There are minor variations in the wording and sometimes variables, but in general if I see certain things it's a strong signal that they used AI.

I added some additional things (see my original comment) and it correlates perfectly with my observations.

Re: When you ask ChatGPT "Tell me a story" it's always is about a girl named Elara

#70
post #26

Earlier quoted context omitted.

Why do people write these kinds of "answers" that the model gives. It's not like the model knows why it's doing anything.

Not understanding how LLMs work.

Ah such condescension! I was careful not to provide an opinion, but replicate the answer as is. The intent was not to treat that answer as fact; but I thought that response was pretty revealing and in fact supported the parent comment on LLMs having been trained on copyrighted materials. The response chatgpt provided was that OpenAI models are designed to create original content "without directly referencing copyrighted characters". If it was creating original content it needn't have referred to the constraint with respect to avoiding directly referencing the copyrighted characters.
Post reply on HN