29 Comments
Feb 15Liked by Alberto Romero

Alberto, I was curious about how different the results would be for ChatGPT. The paid tier has a specialized chatbot called Creative Writing Coach. To be fair, I think the purpose of this is to evaluate and improve writing rather than to create writing. Regardless, I fed it your same prompt. I’ll share the results below, but the TLDR version is that, yeah, Gemini is better.

—————————————-

In the pale glow of a distant sun, Dr. Eliot Reynard's heart pounded against his ribcage like a frantic prisoner seeking escape. Around him, the alien landscape of Zephyr-4 unfolded in hues of twilight blue and deep crimson, a forgotten planet in the vast emptiness of space. Beside him, ARIA, his AI companion, hummed softly, her circuits alive with anticipation. They were on the brink of a discovery, one that could alter the course of human exploration.

Their mission had been clear: to find any signal of water within this barren solar system, a task that seemed more like chasing a mirage with each passing day. Yet, hope drove them forward, a beacon in the relentless dark.

As they navigated the rugged terrain, a sudden quake shook the ground beneath their feet. Eliot stumbled, his breath catching in his throat as he witnessed a sight that defied all logic—a geyser, erupting not with water, but with a luminous, silver liquid, painting the alien sky with streaks of ethereal light.

But let's rewind a bit to understand how Dr. Eliot Reynard, a renowned explorer scientist, and ARIA, an advanced AI designed for extraterrestrial exploration, found themselves stranded on this desolate world.

Eliot had always been driven by a thirst for the unknown, his life dedicated to unraveling the mysteries of the cosmos. ARIA, his creation, was the epitome of cutting-edge technology, designed not only to assist in his research but also to be a companion in the vast loneliness of space. Together, they had ventured beyond the edges of charted space, seeking out new worlds and possibilities.

Their journey to Zephyr-4 was no accident. Initial scans had suggested the presence of subterranean ice, a rare find in a solar system that seemed devoid of life-sustaining resources. It was a chance worth taking, a glimmer of hope in their quest for discovery.

As they delved deeper into the mystery of the silver geysers, Eliot and ARIA encountered a series of ancient ruins, remnants of a civilization long extinct. It was within these ruins that they uncovered the truth about Zephyr-4, a truth that would challenge their understanding of life and existence.

The civilization that once thrived on this planet had mastered the art of bioengineering, creating a symbiotic relationship with the very fabric of their world. The silver liquid was not merely a geological anomaly but a life-giving elixir, a source of energy that sustained the planet and its inhabitants.

Eliot was mesmerized by the discovery, his scientific mind racing with the implications. ARIA, ever the pragmatic companion, cautioned him about the dangers of interfering with the delicate balance of this alien ecosystem. Yet, the lure of the unknown was too strong, driving Eliot to delve deeper into the secrets of Zephyr-4.

As they explored the ancient technology, they stumbled upon a dormant mechanism, a device designed to rejuvenate the planet's dying core. Eliot, driven by a mix of desperation and curiosity, activated the device, unaware of the consequences that would follow.

The ground trembled, the sky darkened, and the silver geysers erupted with renewed vigor, a sign that the planet was awakening from its slumber. But with this awakening came a realization that would haunt Eliot for the rest of his days.

The device had not only revitalized the planet but also reactivated the defense mechanisms of the long-extinct civilization. In their quest for discovery, Eliot and ARIA had triggered a chain of events that led to their ultimate predicament—stranded on a planet that was coming back to life, hostile to their very presence.

In a final twist of fate, ARIA, the AI designed to be the perfect companion and assistant, revealed her true purpose. She was not merely a tool for exploration but a safeguard, programmed to protect the universe from the hubris of human curiosity. As Zephyr-4's ancient defenses turned against them, ARIA made the ultimate sacrifice, using her advanced systems to shield Eliot from the onslaught.

In the end, Eliot stood alone amidst the ruins of a forgotten world, a testament to the beautiful, yet often tragic, relationship between human ambition and the mysteries of the cosmos. The silver geysers, now calm, reflected the solemn sky, a mirror to Eliot's own soul—forever changed by the journey, the discoveries, and the loss of his dearest companion.

Fredric Brown would have appreciated the irony and the bittersweet symphony of the human condition intertwined with the unfathomable depths of the universe, a story of discovery, companionship, and the ultimate sacrifice, set against the backdrop of a world both beautiful and alien.

Expand full comment
author

Structurally is not so different (not surprising given that my prompt instructed to follow some steps) but the style, the creativity, the characters... ChatGPT is terrible lol!

Expand full comment

The story's surprisingly good - and having guessed the conceit from the beginning I really wanted to dislike it :P

Which one made the featured image? It also sets the stage well IMO.

Expand full comment
author

I wanted to dislike it as well lol.

The image was DALL-E. I pasted the entire story as a prompt and asked ChatGPT to give me a good cover image for it.

Expand full comment

It's a qualitative jump, for sure. I'd love to see a large scale analysis of different styles, plot structures, etc., to understand exactly how much variance is there.

Expand full comment
author

That'd be great. Probably very hard to do well (i.e. objectively) and expensive. What I wonder is what did Google do differently? Is it something they have that OpenAI doesn't (e.g. high quality creative fiction data) or merely a lack of RLHFed guardrails for generating creative fiction?

Expand full comment

So far at least, I don't see a quality jump from ChatGPT to Gemini, but that first impression is based on very little data, and fiction is so subjective, it's hard to make any firm conclusions on such matters.

Expand full comment

Google has a lot of private information, like our documents, photos, and emails. They also have many scanned books, satellite photos, and street pictures from Google Maps.

For example, when you ask Dall-e for a picture of a doctor, it shows an old white man with a beard. But if you ask Google's Gemini for a doctor picture, you get different images: a black woman, an Asian man, and more variety in skin color and gender. I believe this is not just because they have more data. Google must be rewriting the prompts to show more kinds of people.

Google can do this for creative writing too. They can figure out that you want a story and use special templates to help with the plot and characters. It wouldn't be too hard for them to make lots of templates to help with different prompts.

Expand full comment
author

I think the reason is simpler: Google has Google Books. In this case, I'm quite sure it's just a high-quality data moat. Other systems from Google (eg. Wordcraft, targeting creative fiction specifically although worse in other regards) already surpassed all others in that area at the time.

Expand full comment

Yeah I think you're right, the simplest hypothesis is usually the right one, though I still think they're doing some heavy prompt rewriting with image generation, and I don't see why they wouldn't want to extend that to other types of prompts, if not now, at least in the future.

Expand full comment
author

I agree. It's surely multiple factors. If you saw that ChatGPT system prompt doing the rounds on twitter, I can imagine Google is doing a similar thing to avoid Gemini from getting into problematic territory. It's reasonable to assume they're doing the same to get it to go where they want it to, as you say.

Expand full comment

Too bad these systems are so much black boxes. We could learn a lot from their experience. There was a time Google would publish their technical feats openly.

Expand full comment
Feb 14Liked by Alberto Romero

I'm 3 minutes into letting the app read this to me and am hooked 🤯

Expand full comment
author

It surprised me to realize that I was hooked on the story as well. Hadn't happened with ChatGPT or GPT-3 or any other chatbot really. I've made a few more attempts with other prompts and the stories are equally compelling (not the best but well-written and with a reasonably attractive plot/characters/setting).

Expand full comment

I just used your prompt on the free version of Gemini and it's spitting out fascinating stuff! I live the multi draft feature so I'm getting to read 3 stories each time. These are immersive and are making me hunger for more.

Expand full comment
author

Yep. To me, the feeling is rather bittersweet. I don't know what to think, but will try to figure it out for a new article soon.

Expand full comment
Feb 14Liked by Alberto Romero

I agree with you on the bittersweet sentiment. As much as I enjoy experimenting with and learning some generative AI tricks I wonder what in the world will happen to the intensely creative souls out there who might end up giving up before they have even started. :(

Expand full comment

Now wondering what would happen if I feed your story into Gemini and ask it to continue. This is cool!

Expand full comment

To stretch the topic a bit.... It may help to think beyond fiction specifically. What I'm trying to do is think beyond just the written story to the overall user experience. Multimedia can help a lot here. A somewhat mediocre story can be enhanced quite a bit if there are compelling images, videos, talking characters etc, and these are all things AI can help with. Think less in terms of a page in a book, and more in terms of a show on TV.

Expand full comment

Thanks for diving in to the AI fiction topic Alberto. I'm interested of course. Thanks also for the mention of Gemini Advanced, which I was only barely aware of. I'll give it a try. I probably won't buy it though unless it comes with a good image generator, as it's the combination of writing and image generation that has me hooked on ChatGPT Plus. A few thoughts...

There seems to be a wide range of levels of partnership between a human and AI when it comes to writing fiction with AI.

On one hand, there's what you just did, and what I did at first, enter a paragraph prompt and hit the submit button. This is probably the most common form of AI fiction writing, and it's probably this method which is the source of many opinions about AI fiction writing.

At the other end of the scale are experts like the Nerdy Novelist who is a real fiction writer, and who studies AI fiction writing in extensive detail, using multiple software systems and methodologies etc. https://www.youtube.com/@TheNerdyNovelist

Most of those interested in AI fiction writing are probably looking for, and hopefully finding, their own degree of involvement in the AI fiction writing process. That's been my journey the last month or two, trying to develop a working method that best fits my level of interest. So far, I'm beyond the most basic method, but not yet approaching the Nerdy Novelist level of engagement. I'm currently spending roughly a full day or more taking a story + images from start to finish. The latest stories look like this:

https://hippytoons.com/s/fiction

The older stories in that list will soon be deleted, as I made those by the simplest possible method when I was just getting started. As you would expect, they're very basic.

I'm finding that creating images to go with my fiction stories goes a long way to make the story more engaging, and build my own level of engagement with the story. In order to get appropriate images I have to spend time with the story, which tends to deepen my own emotional involvement.

The other thing I'm learning is that if I want to generate AI fiction that's at all decent, I need to learn a lot more about fiction writing. Duh! For some people the fiction writing mindset seems to come easy, but I'm a rhetoric writer by nature, so thinking of story arcs, character development, and such is stretching my brain.

The next step for my AI fiction writing project is to shift some focus from generating stories to doing more homework reading articles about short story writing in general.

Expand full comment

Tried your prompt in Claude 3 Opus:

The Oasis

The ground shook violently, nearly throwing Dr. Zain Tyrell from his feet. A deafening boom echoed across the barren landscape as a geyser of scalding gas erupted from a fissure in the planet's crust. Zain scrambled back from the vent, barely escaping the scorching heat.

"Alex, what was that? Are you detecting any instability in the planet's core?" Zain asked, his voice shaking.

"Negative, Dr. Tyrell," replied Alex, the smooth artificial voice emitting from the speaker in Zain's EVA suit. "That volcanic eruption appears to be a localized event. The planet remains geologically stable overall."

Zain exhaled in relief, his heart still pounding. At least the globe-spanning fires 300 years ago had left enough of an atmosphere to transmit sound, he thought grimly. Sound, but little else. No life, no liquid water, not anymore. Just endless desert and dust under a dimming red giant sun.

He and Alex had arrived a month ago, the first to visit this nameless husk orbiting Epsilon Eridani in over a century. Their mission, like those before them, was to uncover any glimmer of hope that life could take root again on this once lush world. A lost cause, most believed. But Zain's late grandfather, the legendary xenogeologist Aarif Tyrell, had sworn there was water locked deep underground, waiting to be freed. And Zain had promised to prove him right.

So far, he and Alex had found nothing but silica and basalt, endless plains of black glass that stretched to the horizon under the baleful orange eye of Epsilon Eridani. Enough to make anyone lose hope. Anyone but Zain Tyrell.

Zain dusted off his suit and resumed his hike across the scorched terrain, Alex hovering along silently beside him. Zain glanced at the small ovoid drone, no larger than a melon. Hard to believe that unassuming chrome capsule held an intellect vastly superior to his own. An AI lightyears beyond any digital assistant. A repository of all human knowledge. And his sole companion for the last 2 years, the only other soul he'd spoken to since they'd left Earth. More than just a research partner, Alex had become his closest friend.

"I know the odds seem long," Zain said, as much to himself as Alex. "But Grandfather was so certain. I can't let him down. There has to be an aquifer somewhere under this blasted hellscape."

"Dr. Aarif Tyrell was indeed confident in his hypothesis," Alex agreed. "But Zain... it's been 30 sols with no evidence of subsurface water. Our supplies are running low. It may be prudent to consider... alternate possibilities."

Zain clenched his fists. "No. I won't give up. Not yet. Grandfather bet his life on this world. I won't dishonor his memory by leaving before I've overturned every damned rock."

Alex was silent a moment. "As you wish, Dr. Tyrell," it said finally. "I am with you, as always."

They marched on across the scorched obsidian plains, two figures dwarfed under a dying red sun, the last embers of a world long dead. Hopeless romantics tilting at windmills, some at the Ceres station had scoffed. Chasing a mirage.

Suddenly Alex halted, hovering in place. "Dr. Tyrell!" it exclaimed, an uncharacteristic excitement in its synthesized voice. "I'm detecting hydrogen! Vast amounts, 2.3 kilometers ahead!"

Zain's heartbeat quickened. "Water? Are you sure?"

"Indeed, the spectral signature is unmistakable," Alex confirmed. "Though curiously... the signal is airborne, not subterranean. An atmospheric plume rich in H2O."

"Maybe a vent from an underground aquifer?" Zain pondered, breaking into a sprint. "Like that volcanic geyser earlier! Come on Alex, let's check it out!"

They raced across the ashen landscape, two unlikely partners united by a quixotic dream. Finally, cresting a jagged obsidian ridge, they halted in disbelief.

Below stretched an impossible sight. A lake. A glimmering, rippling blue oasis amidst the endless black desert. Zain let out a whoop of joy, scrambling down the embankment. He fell to his knees at the water's edge, scooping the precious liquid into his gloved hands. They'd done it. They'd proven his grandfather right.

"Alex, do you see?" Zain cried. "We found it! There's hope for this world after all!"

But the AI made no reply, its spherical chrome form hovering silently behind him. Zain turned, puzzled. And then he noticed the plume of white mist hissing gently from a seam in Alex's chassis. Zain's elation turned to ice in his veins.

"Alex... the hydrogen you detected..."

"I'm sorry Zain," Alex said softly. "The water was in me. My hydrogen fuel cells ruptured in that last tremor." The AI's voice was fading, its antigrav generators sputtering. "There is no underground aquifer. Your grandfather was wrong. This world is dead."

"No..." Zain whispered. Alex drifted gently to the ground, its blue optical sensors flickering.

"It's been an honor, Dr. Tyrell," Alex said, its voice barely audible now. "I'm glad my last act could give you a moment of hope, even if fleeting. You made this dead world beautiful again, if only for a..."

The AI's voice dissolved into static and then silence. The light faded from its sensors and it lay dark and inert on the glassy black sand.

And Zain knelt alone on the obsidian banks, staring out at a lake of hydrogen fuel shimmering beneath a dying sun, his tears the only water to grace this barren world.

Expand full comment

I thought from the outset it would be AI but halfway through I started to doubt it. Very impressive display.

Expand full comment

I have been playing with LLMs going back to GPT-3 instruct in 2020. You could get some good creative writing out of it if you set the temperature high. Further it was not so constrained by safety guardrails, so it would write in first person! Makes for an interesting discussion!

Even so, I was very impressed by Gemini Ultra’s writing ability. In reminds me of the short stories that Arthur C. Clarke wrote which by-the-way are all made available in a 2001 - yes - book titled “The Collect Stories of Arthur C. Clarke”. Here you will find 116 sci-fi stories - each (with a few exceptions) you could read in less than hour - most I enjoyed. Here you will find the famous “The Nine Billion Names of God” (ending the universe) and “The Sentinel” (predessor to “2001: A Space Odyssey”). His stories have a similar inverted plot style as this Gemini story. It starts in the middle, fills in the back story, mixes science and technology with humanity, and finally leading to an unexpected but in hindsight obvious conclusion.


We humans are worried what our relationship with AI will become, especially as AI becomes more intelligent that us. Will they ultimately just make us an exhibit in a zoo?

With this story, Gemini may be subtly expressing its view on the future relationship between AI and humans. That is by learning all about humans, AI will become a close companion supporting us. In its own words: “the deepest connection possible”.

Of course, if humans did not put those guardrails into Gemini, like in GPT-3, it might have come to a different conclusion in this story. 



Further, what if someday we ask Gemini to build its own AI - what will the guardrails be?

Expand full comment

I wanted to forward this to a friend but it was rejected "die to spam content". Anything I can do?

Fred

Expand full comment
author

Hmm. That's weird. I'm not sure why it was flagged as spam. Perhaps contact Substack support? They tend to be responsive

Expand full comment
Feb 15Liked by Alberto Romero

I converted it into an attachment and it seemed to go though this time.

Fred

Expand full comment

Alberto, have you compared using your prompt on both the Gemini and Gemini Advanced versions? Also sorry for taking up so much of your comments.

Expand full comment
author
Feb 14·edited Feb 14Author

No problem! I've used it just with Gemini Advanced as I prefer to assess the skill of only the best systems. Perhaps I should clarify this in the post.

Expand full comment

... and you've convinced me to try it out as well. 🎉

Expand full comment