š® Exponential View #552: Three years of ChatGPT. āInvisibleā productivity. DeepSeek uncensored, the bubble watch, GLP-1s++
An insider's guide to AI and exponential technologies
Hi all,
ChatGPT launched three years ago on this day. In that short time, access to āgood enoughā intelligence has grown massively.
Iāve spent a decade analysing the exponential age (I wrote a book about it in 2021), but these three years mark a distinct turn with overlapping S-curves in models and chips, new behaviours, and a growing gap between exponential technology and linear institutions.
A four-part series sets out my current view of AI as a general-purpose technology at this inflection point. I cover how I see it reshaping firms, markets and the wider economy:
Part 4 will be in your inbox today, right after the Sunday edition!
An invisible flood of productivity
Anthropicās new study says that even if AI progress stopped today, existing tools could lift US labor productivity by about 1.8% a year for the next decade ā a historically unprecedented pace compared with the postāwar boom or the dotācom era. I agree with this sentiment ā in fact, I think itās probably been true since GPT-4. The tools are powerful but implementation is tougher: companies need to go through wide-scale people-centric transformation processes to make best use of them. This takes time.
Anthropicās projection matches what many economists expect. But weāre not seeing it yet: recent US productivity growth is around 1.5%, roughly the historical average.
Although visible AI use is clustered in tech hubs and roles like software development, much larger āsubmergedā exposure exists across admin, finance, and professional services and makes up 11.7% of wage value, five times higher than the visible tech sector (see MITās Project Iceberg for details on this research). For this reason, the biggest impacts will emerge outside classic tech centers.
So far, the impact isnāt showing up clearly in the broad stats ā and itāll matter that it does in 2026.
Are we in a bubble? A weekly update
š Quick read: Not a bubble yet. Application-layer revenue is strengthening, but funding quality continues to show signs of strain.
Three key movers this week:
Nvidia has issued a detailed memo to analysts systematically refuting bear cases regarding circular financing and accounting irregularities. It came across as a little desperate.