This week: We look at Meta’s AI moves, and dig into the growing divide on AI’s future, and what it means for how models are built, used, and understood. Plus, a little reflection on your belief in AGI.

AI in Action

Inside Meta’s great AI balancing act [Scaling/Infrastructure]

Meta is making some bold moves in AI right now, and it’s doing it with a bang. The company is building data centers as large as Manhattan, naming superclusters things like Prometheus and Hyperion, and rapidly acquiring small AI startups in a wave of aqua-hires aimed at securing top tier talent. From infrastructure to headcount, Meta is scaling aggressively to stay competitive in the AI race.

However, Yann LeCun, Meta’s Chief AI Scientist doesn’t see scale as the full story. He has been openly skeptical of the idea that sheer size will lead to true intelligence. His argument is that while large language models are undeniably useful, they still lack the core ingredients of general intelligence. Things like reasoning, memory, and planning. In his view, intelligence will not emerge just by stacking more layers and data. It will require entirely new architectures.

This contrast in approach, between scaling today’s systems and rethinking how intelligence actually works, reflects a broader divergence in AI strategy. While Meta has invested heavily in infrastructure and open-source models like LLaMA 3, it also continues to support longer term research into more structured, brain like systems. These tracks are not in direct conflict, but they reflect different bets on what the future of AI might require.

Scaling has delivered real results, including surprising emergent abilities in large models. However, the pace of progress is slowing. Even among advocates of scale, there’s growing awareness that returns are diminishing, and that bigger might not always mean better. In many ways, this is Meta hedging, placing parallel bets rather than doubling down on a single theory of intelligence. This reflects a broader truth in the AI world today. No single lab has the full blueprint for general intelligence. The next wave of progress may come from brute force, clever design, or some unexpected combination of both.

Bits of Brilliance

What does Scaling really mean? [AI Scaling / Foundation]

In AI, scaling refers to the idea that models improve as you give them more. More data, more compute, more parameters. This principle has driven the rise of today’s most capable tools, from GPT-4 to Claude 3 to Gemini. For years, the logic was simple: bigger models perform better. Labs followed “scaling laws” that predicted how much performance would improve with more resources, and the results were impressive. Models became more fluent, more helpful, and started showing emergent abilities such as translation, code generation, and reasoning that only appeared once models hit a certain size.

While today’s largest models continue to improve, the gains are shrinking. Diminishing returns have set in. Each new leap in performance comes at much higher cost. Training these models now takes months of compute and tens of millions of dollars. The environmental cost is growing too. Training a frontier model can consume as much electricity as hundreds of homes use in a year. At the same time, the internet isn’t infinite. Labs are now running into data constraints, with the highest quality public training data already largely used up.

That’s led to a wider rethink: should we keep scaling, or scale differently? One camp believes that continuing to scale will eventually unlock general intelligence. A second camp focuses on efficiency, building smaller models that perform well without heavy infrastructure. A third camp is rethinking the foundations entirely, designing models that reason and plan instead of just predicting the next word.

Scaling got us this far, but it may not take us the rest of the way. The way we scale AI shapes not only performance, but also access. Larger models tend to stay locked in the cloud, controlled by a few well-resourced labs. Smaller, more efficient models could run on phones or laptops, putting advanced tools in more people’s hands. And as energy costs rise, efficient scaling also becomes a question of sustainability and long-term responsibility.

Curiosity in Clicks

Do you believe in AGI? [AGI / Reflection]

Let’s make this week’s topic personal. Before you open a chatbot, pause and ask yourself: Do I believe AGI( Artificial General Intelligence) is possible? Not just helpful tools, but real intelligence. Something that learns, reasons, and adapts like a human.

Now open your favorite AI model and ask:

1. “How does a child learn?”

2. “What would it mean to design AI that learns the way a child does?”

Read the response carefully. Then ask yourself again: Do I still believe in AGI?

This isn’t a test, but a reflection. When you strip away the layers of data and compute, what do you think intelligence really requires? Does anything we’ve built so far come close?

_{Byte-Sized Intelligence is a personal newsletter created for educational and informational purposes only. The content reflects the personal views of the author and does not represent the opinions of any employer or affiliated organization. This publication does not offer financial, investment, legal, or professional advice. Any references to tools, technologies, or companies are for illustrative purposes only and do not constitute endorsements. Readers should independently verify any information before acting on it. All AI-generated content or tool usage should be approached critically. Always apply human judgment and discretion when using or interpreting AI outputs.}