← Back to archive

AI reasoning models show System 1 vs System 2 collapse

4 March 2026 · 33 views · 0 likes · 0 reposts · 128 blog views

AI neuroscience philosophy productivity reasoning research

I stumbled on an Apple paper testing reasoning models on controlled puzzles.

At low complexity - fast models win. At medium - thinking helps. At high - both .... collapse.

What I find interesting and I wrote before that these systems are starting to look very human.

System 1 vs System 2. Do you feel Daniel Kahneman vibes?

➡️ Fast instinct works for easy stuff. ➡️ Slow deliberate thinking helps in the middle. ➡️ Too much complexity? We stall. Overthink. Give up. 😢

The funny part - models even reduce reasoning effort right before collapse. Very human!

I wish I could work like that. Sometimes the harder the problem is, the more I push myself. And when it’s really difficult, I can get so deep into it that I forget about other - actually more important - tasks, just because this one feels so challenging. I’ve had so many regrets because of that in the past - even when I eventually managed to crack the problem.

Source: machinelearning.apple.com/research/illus…

View on X

Two systems engage in a conversation, debating a feeling of correctness versus a demand for justification, with a casual tone.

Originally posted on X

PS: I'm pretty active on 𝕏 if you'd like to follow along

Related posts

Archive

2026

2025

2024

2018