May 15, 2026
Curve drama goes full S-mode
The sigmoids won't save you
Fans roast the "it’ll flatten out" crowd as AI prediction fights get very messy
TLDR: The article says people keep wrongly assuming fast AI progress must start slowing soon, even though similar predictions have failed before. Commenters split between mocking the article as obvious math, questioning the charts entirely, and arguing the real limit may just be lower than the hype suggests.
The big argument here is surprisingly simple: just because fast growth eventually slows down doesn’t mean it’s about to slow down right now. The article takes aim at the smug internet line that "all exponentials become sigmoids" — basically, the idea that every boom has to flatten out soon — and says that this is often used like a conversation-ending magic spell in AI debates. The writer brings receipts from birthrate forecasts, solar power predictions, and one very awkward AI forecast miss, where a careful-looking curve got blown up by the next model release.
And the comments? Absolute peanut-gallery energy. One camp shrugged and said, yes, okay, but these curves still did explain earlier AI progress just fine. Another camp went full professor-mode, reducing the whole essay to: the early part of an S-curve doesn’t tell you much, thanks for coming to my TED Talk. Then came the skeptics asking the question that always starts a fight: what even is this graph measuring? One commenter flat-out said they weren’t buying that a newer model had "double" the capability of another, while another insisted that if you just changed the chart scale, the prediction didn’t look so bad after all.
The funniest part is the vibe: half the crowd is yelling "you called the slowdown way too early", while the other half is muttering "maybe the ceiling is lower than you think anyway". It’s less a neat math debate and more a classic internet brawl over whether the future is overrated, mismeasured, or arriving faster than anyone wants to admit.
Key Points
- •The article argues that stating exponential trends eventually become sigmoids does not determine when AI capability growth will slow.
- •It uses epidemic spread as an example of a genuinely sigmoidal process driven by limits in the susceptible population.
- •It describes airspeed-record progress as occurring across multiple technology generations, with ramjets reportedly plateauing around 3500 km/h due to technical and economic constraints.
- •It cites UN birthrate projections, solar power deployment forecasts, and a Wharton analysis of the METR AI capabilities graph as examples of forecasts that predicted flattening too early.
- •The article says that forecasting when growth slows requires understanding the underlying mechanism generating the trend rather than relying on the general fact that growth cannot continue forever.