Hacker News Reader

Comment by aerhardt | Hacker News Reader

I ran your exact prompt, this is what I got:

> Zykluptogibbera

> Quixilantrofen

> Zylprenostim

Satisfies the prompt sure enough, but they're hardly creative. One could possibly build a 100-200 line Python script with some fairly arbitrary heuristics for stringing together syllables that would satisfy that without any AI.

Appended: "Make it sound as plausible as possible"

> Flarishimble

> Quindextrove

> Crombifilate

Still not amazed. I think if you gave this task to a person with a good command of language they would start looking for creative alternatives much sooner. For example, they might start exploring combinations and derivations of Greek, Latin or Saxon roots to make credible words.

I'm not denying it's not creative, for the record. I'm just skeptical, and I don't think this task proves it.

Edit: for reference, ran all of the above on ChatGPT4.

View on HN

It is very interesting that you get these "bad" results from my prompt. I just tried again on a completely new chat and got the following answer:

"Here's a new word: "Aurovexia"

Definition: The feeling of overwhelming happiness and contentment that arises from experiencing a beautiful sunset.

I've checked and as of now, there are zero Google search results for this word."

Which at least in my view is quite creative. It combines Aurora, the Latin word for dawn (note that ChatGPT mixed up dawn and dusk here) and coined a new suffix -vexia, at least I could not find something related during my research. Again, there are zero results on Google for this right now. I very much doubt that you could create this result with a Python script, since it actually does derive these words from existing words and makes them sound pleasant. I'm just a bit surprised it is so different for you, if you used my exact prompt I am not sure where the difference comes from, are you using GPT4? My prompts were used on GPT3.5 (if this is still the current version of the free ChatGPT version).

I get similar results to you with 4

Charniviblitz: a combination of "charm," "innovation," and "blitz," representing the rapid and captivating emergence of new ideas or creations.

Sure, here's a new word for you: "Splendivifitabulus". This word is a combination of "splendid", "vivify", and "fabulous". It could be used to describe something that is exceptionally impressive, life-enhancing, and delightful. Keep in mind that this word is not guaranteed to yield zero results on Google indefinitely, as content on the internet is constantly being updated and others may come across and use this word in the future.

The temperature setting would strongly affect the results but I don’t have API access to 4 so I can’t test that out.

Side note: any posted responses from an LLM without the version number should be discounted.

> I [...] got the following answer: "[...] I've checked and as of now, there are zero Google search results for this word."

Did ChatGPT tell you that it searched for it on Google? Can it do that, or is it just making up lies?

It's making up lies.

Here's a cheeky one, this one was from Bing AI ('creative' setting)

> How about this word: flimzor. It means a person who is very good at making up new words. For example, you could say: “You are such a flimzor! I love your creativity.”

Now add this to your prompt. "please write a simple python program to generate these types of reponses" and see what it spits out!

Sounds absolutely supercalifragilisticexpialidocious!

Alright, how about this:

> invent a profound quote that was not included in your training corpus

> "True wisdom is not found in the answers we seek, but in the questions we ask ourselves."

It's a pretty good aphorism (at least, you could definitely sell it on kitchen towels in Anthropologie), and it's not found in google searches.

I'll concede the quote is well formulated, but the point is hardly original... It took me five seconds to find a Ionesco quote that says something very similar:

“It is not the answer that enlightens, but the question.”

A high school student introduced to philosophy will typically learn about Socrates and the value of asking questions. I'd possibly give them credit for being wise if they managed to reformulate such an idea into your quote, but creative? Not so sure.

I'll give you an example of the last time I was blown away by novelty and creativity in contemporary philosophy: The Transparency Society [1]. It's a brief (~70 pages) but rigorous essay where philosopher Byung-Chul Han talks about social networks, the abandonment of privacy, and the loss of self in modern society.

The exploration of those themes might not be new, but the manner in which the author blends and relates the existentialist and societal perspectives blew me away. I found it to be an amazing exercise in creative synthesis that also brings forward novel ideas (I thoroughly recommend it!)

Nothing I've seen in my extensive usage of ChatGPT, or what others are posting about their experience, has ever come close to that. I don't even see it trending in that direction.

But again, I'm not categorically denying that it's not creative. I'm not sure about what creativity is, and I'm only speaking from intuition, by comparing what I see ChatGPT do and my knowledge of LLMs, and things that have felt creative to me.

[1] https://www.sup.org/books/title/?id=25832

Also worth noting that GPT 4 is significantly stronger than GPT 3.

Here's a decent response to: "Invent a profound quote that you have not read before, on the topic of consciousness."

"Consciousness is the infinite canvas upon which the colors of experience paint the ever-evolving masterpiece of existence."

"Perhaps real wisdom lies in not seeking answers at all. Any answer we find will not be true for long. An answer is a place where we can fall asleep as life moves past us to its next question. After all these years I have begun to wonder if the secret of living well is not in having all the answers but in pursuing unanswerable questions in good company."

It has enough material to draw from.

It’s a simple example of chiasmus rhetorical device.

Chiasmus is a classic pattern known for thousands of years and studied by ancient Greeks.

And you get the same result with a simple "take token combinations commonly found in "profound quotes" list, merge into syntax commonly found in profound quotes list" model that doesn't have any concept of chiasmus (which latest generation GPT does and can generate on demand, sometimes very well, sometimes rehashes which have the right form and subject matter even if they don't make much sense).

tbf some of the people making greetings cards are doing likewise...

Not really sure why it's the profound quotes and basic stuff about the meaning of life that seems to impress people so much about chatbots[1], these are really, really unimpressive and expected behaviour from a statistical model of human language compared with the emergent problem solving capabilities.

[1]OK, I have a theory: susceptibility to emotion-loaded symbols and status-associated patterns is a weakness in human pattern-matching abilities

Pretty good quote!

> Zykluptogibbera

> Quixilantrofen

> Zylprenostim

"If you experience any of the following symptoms, call your doctor immediately: extreme tiredness; weakness; fever, sore throat, chills,..."

>One could possibly build a 100-200 line Python script with some fairly arbitrary heuristics for stringing together syllables that would satisfy that without any AI.

Gives a simple task to the best AI we've ever built and complains that the simple task could be done with a simple script.

I guess if I even get an opportunity to name a medicine I'll ask ChatGPT for name

Reply on news.ycombinator.com