The Flow State and AI
There’s a sweet spot where humans hum like machines - the flow state . One task. Full focus. Vivid Imagination. A few hours of unbroken attention—and the whole system locks into flow. Time blurs, output spikes, meaning returns.
Step outside that zone and we look like an AI hallucinating - our words and actions have no coherence to our stated goals. We doom-scroll, tab-hop, snack on sugar, binge on noise. Our stated goal might be lean and muscular or create something meaningful, yet we burn the same precious tokens—minutes and calories—on junk.
If a large-language model drifted this far off prompt, we’d call it a hallucination. When we do it, we call it distractions. Humans being operate best in flow state - singular focussed on one task for a few hours is usually how you achieve flow state.
When the brain flags low battery we often grab the nearest charger, smart or junk. That’s a separate topic. Today’s fight is different: building an AI that stay locked in flow longer than we can.
I understand that our flow state and AI hallucinations are different phenomenon, but its interesting to think about them in a similar way.
But the current AI hits the zone for 10-15 glorious minutes before getting sloppier than . The context window floods, weights spreads across large amounts of text , and micro-errors snowball. The model mirrors our own slips too: sloppier prompts over time.
We can patch some of this leak with RAG, longer context windows, prompt engineering—but I don’t think it is a long term answer.
In order for AI to be in flow-state, we need a more foundational architecture shift that remembers without bloating, stay razor-focused for hours without guzzling teraflops like current CoT or ToT methods do, then we can enter a new AI era: close to zero-hallucination helpers that out-concentrate the monks.
Maybe the AI isn’t ever meant to be perfect, just like us.
Also, maybe we also spin up future models that embrace imperfection, trading accuracy for lightning strikes of creativity—Picasso AI alongside accuracy demanding current wave of AI. It can be a true creative with spurts of inspiration and a vivid imagination, wilder and crazier than ours.
Either way, waiting for the day when a team of researchers, unknown today, create the next breakthrough.
