Farcaster

🧠 Apple says your favorite AI isn’t “thinking” — it’s just guessing smarter

In a new study, Apple researchers claim today’s reasoning models show an “illusion of thinking” — they look smart, but fall apart once tasks hit real complexity. Performance? Drops to zero, even when compute isn’t the problem.

🧩 Using puzzle setups like Tower of Hanoi and River Crossing, they found:
– Easy = fine.
– Medium = decent.
– Hard = complete collapse.
Models start reasoning… then suddenly stop trying — cutting token usage right when logic is most needed.

🍏 Apple calls it a pattern-matching illusion — complex language, surface-level smarts. Even with clear solution steps, the AIs flopped hard on tricky cases.
🔥 This drops just before WWDC — accidental timing or subtle shade? Either way, message is clear: AGI isn’t “thinking” yet — it’s still pretending.