While there’s always value in exploration, frontier economics heavily favor second movers, so don’t let yourself be whipped into AI FOMO.

GenAI is an exciting new frontier and has consumed basically all of societies attention and money.

This comes with a sense of importance, being part of the future, enlightened above the rest and responsible to also ship them into the frontier lest they may be left behind.

This is a myopic view. Zoomed out, we can see of the knowledge will be ephemeral as knowing assembly language or windows for workgroup networking or prompting stable diffusion 1.5 from 4 years ago is today.

Half life of knowledge on the frontier is shorter than ever in the history of mankind, and of the many branching paths taken, only a handful lead to durable long term patterns and learning.

The knowledge gained for sweating the “prompt engineering” craze of 2022, the RAG an Chatbot obsession of 2023, the “cram AI auto complete” into every product surface and “AI IDE” wave of 2024 has mostly expired, with only narrow pieces still holding value today.

Because it’s not a race, as the techbros would like you to believe, but a marathon without end.

There’s value in tinkering but from an energy conservation perspective, the critical strategic choice in marathon, second mover is the right position to take for all but the most ambitious competitors.

Nowhere is this dynamic more apparent than in frontier lab AI training. Of the hundreds of billions of capital burnt to train ephemeral “top tier” models to dominate benchmaxxed charts for 2 weeks, fragments of fragments of value remain … an orgy of infrastructure spend being burnt in glorious GPU fire into NVDA fumes huffed by the entire US economy to keep the GDP afloat.

Meanwhile, on a fraction of that spend, Qwen, DeepSeek, GLM and friends moved on the path cleared by Llamas with superior, orders of magnitude cost efficiency, spending on science, not GPU.

Where is your Sora God now, laughs AI Morgan Freeman on Seedream2, reminding US companies and customers that they still have to pay back investors, that the act of pushing the pedal to the metal has depleted years of capital that could have been spent more wisely on high speed rail, clean energy, poverty reduction, education, healthcare and opportunities in society.

And then there’s the nature of the technology, it’s uncanny ability to automate and commoditize any task of value as long as it has discernible patterns.

By all means, learn, understand, explore, but don’t whip everyone into the frontier s where Darwinism favors the second moving along the paths derisked by others.

Should AI automates all the jobs, and the consumption economy doesn’t collapse, frontier jobs are the first, not the last to go, because they are feeding the very data pipelines training their own replacement as 2022 prompt engineers found out when Dalle-2 and Nanobana killed their magic “best quality, 10 FINGERS ONLY, NOT UGLY!” incantations.