Is AI thinking or just imitating our thinking?

A new Apple research paper, "The Illusion of Thinking," suggests these reasoning models are doing the latter. Sophisticated mimicry at a previously unimaginable scale.

In an AI with Alec fireside chat, I spoke to Henry Saba to gather his perspective on the paper and get his take on what this really means for businesses and builders in the near-term.

As a leader in the space of building data infrastructure, automation and intelligent systems for Citadel and Lazard before starting his own company, Specialized Data Company, Henry’s perspective was insightful and clear.

Exactly the kind that is the most helpful when trying to make sense of the exponential change all around us:

1️⃣ Biggest takeaway from the paper is “their inability to solve generalizable problems”

2️⃣ Think of LLMs as pattern-matching machines (aka massive idioms / Mad Libs → s / o Jack Clark and Rick Rubin) vs LRMs as systems that appear to think through problems step-by-step

3️⃣ Put “AI at the core of the systems that you build” but be sure you’re asking it to solve problems it has seen before

4️⃣ Near-term, “tightly scoped” and “very refined, very purpose-built agents for specific use cases” is where a ton of value will be generated

5️⃣ Gap between AI marketing hype and product / capabilities is getting really wide, critical to know the difference

Research link: https://ml-site.cdn-apple.com/papers/the-illusion-of-thinking.pdf