Theory Thursday: Exponential Growth

Do you remember the trick question: water lilies that double every day. Now the lake is half covered. When will it be fully covered?

At the beginning, exponential growth feels slow. It builds momentum and - seemingly suddenly - arrives with full might. AI is in a period of rapid exponential progress. In several domains, algorithmic efficiency and performance have been improving at rates comparable to Moore’s Law, with some measures doubling every 6-12 months.

But: has AI actually created tangible impact beyond gimmick?

A few examples really caught my eye.

• Robots and AI working together to find a new, less toxic compound for future LEDs (they performed 64 real experiments, improving material performance by around 150%).

• The designed-by-AI drug Rentosertib shows improved lung function in Phase II clinical trials, meaning it could eventually help patients breathe a bit easier.

• AlphaDev discovered a faster way to sort small data sets. The algorithm is now part of LLVM, bringing small but measurable speedups to nearly every modern device.

• AI-based heart-failure detection can work from simple ECGs, making diagnosis more accessible to underserved regions.

It’s looking like a self-amplifying circle: faster algorithms enable better models, which unlock new science, which justifies more compute.

👉 If the doubling continues, what other breakthroughs are just one cycle away?

• Zhou, Y. and Li, F. et al., Autonomous discovery of high-performance polymer mixed conductors through AI-guided experimentation, https://lnkd.in/e8tx37RJ

• Insilico Medicine, Rentosertib, an AI-designed drug improving lung function in Phase II clinical trials, https://lnkd.in/er_WFfkR

• DeepMind, AlphaDev: branchless sorting algorithms merged into LLVM libc++, https://lnkd.in/edR3brBA

• Zhou, J. and Kumar, S. et al., Heart-Failure Detection in Underserved Populations using AI and ECG data, https://lnkd.in/exqtry-h