The Taste of Grok
Today I did a little experiment to see if an LLM (Grok) could understand my own personal reading tastes enough to predict my reading preferences. My dream is that this technology could be trained by individuals to sift the self-published book slush pile that is Amazon for personalised gold nuggets.
To test this idea, I took some time to experiment with Grok using some short stories. Analog’s Analytical Laboratory Finalists became the training set first. I let Grok analyse each story, creating a 100 word summary of the plot and extracting the best and most representative chunk of the prose. Grok also made a ranked list from best to worst story.
I then took the time to read each story, and informed Grok of my own feelings and ranking list.
Here’s my feedback to Grok for the curious:
The handmaiden alchemist is delightful, but its cleverness is very small scale, which is a refreshing change in sci fi but doesnt quite suit my appetite for big ideas. The Dark at the End of the Tunnel had a wonderful core idea, and I can see they tried hard to construct a compelling frame jumping between times and building up to the climax, but the shrug at the end was a little underwhelming after stretching out the build up. It couldn't decide if it was an action movie or a drama, and the blending of the two felt skilful but maybe overambitious in such a short format? The end result felt a bit longer than it needed to be. Tepid war focuses on a topic I think is very interesting and salient to the near future, but I quickly found myself skimming. Drone warfare existing and being a background to life was quickly and effectively introduced, but then the other layer of character focused drama failed to intrigue me enough to get into it. Susan Rose Sees Mars... I remember reading this one about a year ago. It was a really skilful and lovely blend of a sci fi trope (mars colony) with a fresh angle (how human art grapples with the reality of it). It didn't have a stunning new idea at the core, just an obvious implication of that technological advance which I hadn't seen anyone delve into before, let alone do it skilfully. No "aha" moment to justify reading the whole thing, but the execution of the prose was enough to make it enjoyable to experience in full. Music of a Different Sphere kicked me out immediately. Even musicians today know enough physics to have seen that problem lightyears in advance. A great example of how unconvincing core concept cannot be saved by any amount of prose polishing. I don't even believe Mars colonies are feasible in reality, but can accept that starting premise to make stories like Susan Rose interesting. Reflecting on all that, I think I agree that Dark at the End of the Tunnel is number 1 (best premise, solid execution). Susan Rose is my number two, intriguing if mundane premise but beautifully executed. Number 3 would be Tepid War. Highly relevant topic, but the execution didn't grab me. Number 4 would be Handmaiden Alchemist- the writing was a nice reflection of that period. Lowest at number 5 was Music of a Different Sphere- the unconvincing premise torpedoed it for me from the first sentence.
To see how persistent and transferrable this training data was, I then reversed the order of steps, using Asimov’s Readers’ Awards Finalists.
This time I read and ranked the stories first, making notes about my main reasons. Here is my ranked list from best to worst.
1. An Unplanned Hold- Breathtaking! The concept seamlessly blended the supernatural, the sentimental and the civilisational. The prose shone without ever glaring. Total universe in a grain of sand territory and a testament to what short stories can achieve.
2. A Gray Magic- Overwhelmingly beautiful. A meditation on life and death and connection and meaning. The writing reminded me a lot of Dostoyevsky, not the prose, that is just the medium. But the focus and attention and noticing. The sci fi was just a coat of paint, but it enhanced everything about the story itself.
3. Sunsets- This one opened with a finely tuned assault of delightful specificity, and managed to ease that pace of info flow into a comfortable jog the rest of the way. Absurdly fantastical, but in a way that never felt cringey despite the lack of plausibility.
4. Mere Flesh- Solid whimsical set up. Once I got the basic premise I started skimming the padding in the middle until the build up to the final reveal. The concept in the end was kind of a meh for me. I am deeply sceptical of digital brain interfaces, but the idea of a neural implant taking control of a failing organic brain is a little above average plausibility on this front. Writing was solid, but almost too polished, with a kind of Hollywood/scripted tone which I find off putting in written fiction.
5. The Adherence- Whacky and fun surrealist sci fi, but kind of pointless. Solid writing but unremarkable.
Grok analysed the stories independently and guessed I would rank them as follows:
Final Ranking:
"An Unplanned Hold" - Big idea, tight execution, cosmic stakes.
"Mere Flesh" - Bold concept, engaging chaos, family tension.
"Sunsets" - Poetic craft, emotional depth, modest scope.
"A Gray Magic" - Quietly real, limited ambition, slow burn.
"The Adherence" - Vivid but implausible, small-scale trickery.
Position 1, 3 and 5 were correctly guessed. 2 and 4 were swapped, but when I explained my love for novels by Dostoyevsky and Patrick White over Disney and Hollywood scripts Grok seemed confident it would swapped the recommendations in that case.
All in all, I think this is a very impressive result based on an extremely limited training set. All of these stories were the best of the best, so the AI wasn’t leveraging mundane details like grammar errors or incoherence to pick winners.
In the future this end use for LLM technology is likely to advance and refine its capabilities. It cannot analyse multiple novels, but that might just be a matter of time. Where this leads is anybody’s guess. I have yet to see LLMs produce writing from scratch that can challenge the best writers out there. There are solid arguments that it never will, though that also remains to be seen. But a skilled human plus an LLM…
Photorealistic oil paintings were the height of visual art until photography dropped the price of those images to near zero. Interestingly, we now suspect those masters of such painting were using the high technology of the day (lenses and projection) to elevate their art beyond what was possible with unaided hands and eyes.
My bet is that LLMs will propel writers to create works which quills, typewriters, or Microsoft Wordâ„¢ could not.
These new tools will destroy everything we knew and loved.
But with them we will create everything that we will know and love tomorrow.
