One Person Unicorns
Issue 116 · September 8, 2025
Welcome back to Artificial Insights, your weekly dispatch from the near future.
The Ember months are here, and I expect the next stretch to bring an overwhelming mix of AI breakthroughs, hype, and inevitable disappointments.
What feels most valuable right now are the jagged edges, where customization makes all the difference. GPT-5 is remarkable, but its expanded ability to keep more of my chat history in context is what truly transforms it into a reliable partner. Programming in Cursor with code scaffolds is far more productive than tossing vague prompts into the void. Generating images with style references in MidJourney is far more effective than leaving the model to guess aesthetics.
LLMs explore latent space indiscriminately; you gain leverage by constraining them with context.
One striking use of larger context windows has been in personal health. While traveling, a food-triggered toothache dragged on for a week. With no local dentist available, GPT-5 helped me track evolving symptoms, offering a coherent, likely diagnosis that a later dentist visit confirmed. Previous models would have struggled to connect the dots.
For consumer AI, customization will become the stickiest issue. Why switch to another model if doing so means erasing your accumulated context? I don’t think people will.
…
In the coming months we’ll return to weekly newsletter issues alongside a mix of in-person and remote events.
-
Vibe Coding (Amsterdam, Oct 7): I’m co-hosting the third edition of our vibe coding hackathon at the Embassy of the Free Mind during World AI Week. The last two sessions sold out quickly, so reserve your spot now on Luma.
Until next week,
MZ
MZ
Quick Links:
-
Fascinating Reddit deep dive about how LLMs hallucinate.
-
What is Google Nano Banana? An insider view about Gemini 2.5 image generator.
-
Recent The Atlantic episode exploring how AI has become the new battleground line between democratic societies and authoritarian systems.
-
Kevin Kelly on the Trust Quotient (TQ): a speculative riff on how trust functions as currency in technological and social systems.
-
Maths for LLMs: a technical blog pointing you toward the mathematical foundations necessary to grasp how large language models actually work.
-
I Am an AI Hater: a deliberately contrarian rant that throws politeness out the window and unapologetically calls out AI with blunt disdain.
[
The Absent-Minded ProfessorAn E-bike For The Mind
Thank you for being here. As always, these essays are free and publicly available without a paywall. If my writing is valuable to you, please share it with a friend or support me with a paid subscription…
Read more
8 months ago · 19 likes · 6 comments · Josh Brake
The Cracks in the AI Boom (15min)
Use AI where it clearly saves time and money, but keep your core identity, voice, and relationships human.
Automating Coding Through AI Models (28 min)
Michael Truell points to a dramatic but gradual shift: coding won’t disappear, but it will increasingly resemble collaboration with an intelligent partner.
There will be this long messy middle where you will be working with the AI more and more… like a colleague more and more.
Difference, Creativity, and Why LLMs ≠ AGI (58 min)
David Deutsch offers a grounded counter to AI hype: expect faster work and better tools, but don’t mistake them for human-origin explanatory creativity—or for a shortcut to mass-produced genius.
For AI, yes… the rate of progress… will be greatly increased… But AGI… I don't see any reason why having AGIs around is any better than having humans around.
Between Hype and History (1h15 min)
Benedict Evans frames AI as a major platform shift whose contours are uncertain, but whose competitive and behavioral resets are all but inevitable. Thanks Rod!
It seems to me right now you could do like a double blind test of the same prompt given to Grock, Claude, Gemini, Mistral, Deep Seek. I bet most people wouldn't be able to tell which is which.
Newsletter
Follow us for weekly foresight in your inbox.