❓ Are today's LLMs intelligent? For sure - look what they can do! Surely not, se

❓ Are today’s LLMs intelligent? For sure - look what they can do! Surely not, see how miserably they are failing me! - What is the truth?

🔦 Weekend Flashback. One of the amenities of running trails is that you not only get yourself time in beautiful nature, but you can also use it to listen to deep conversations. This weekend I listened to Machine Learning Street Talk (MLST)’s episode with François Chollet. Still digesting it 🙂. Sharing here, because I considers this podcast a very good use of time - no need to run 😉

💡 In the conversation François presents a compelling perspective on intelligence, consciousness, and artificial intelligence that challenges many conventional views in the field. Naming threes things sticked to my mind: 1️⃣ Intelligence Definition: True intelligence is not about memorizing information or accumulating knowledge, but rather about effectively handling new situations. Chollet argues that current Large Language Models (LLMs) possess “near-zero intelligence” despite their impressive capabilities, as they primarily function as sophisticated pattern-matching systems.

2️⃣ The Kaleidoscope Hypothesis: He introduces his theory that the world’s apparent complexity is actually composed of simpler, repeating patterns that combine in various ways. Real intelligence involves identifying and utilizing these fundamental patterns to comprehend novel situations.

3️⃣ Consciousness Perspective: Consciousness develops progressively rather than appearing suddenly. It exists on a spectrum, with animals possessing varying degrees of consciousness, and human consciousness fluctuating based on circumstances and activities.

His view is that intelligence itself is merely a tool for converting information into useful models, with the critical factor being how humans choose to implement it.

Very inspiring thoughts and deep insights. Brings me back to the pragmatic view that LLMs are fantastic tools, which can bring us tremendously useful result if operated in an intelligent way 🙂 .

What are your takeaways?

Link to episode: https://lnkd.in/eCaKCzHD

---