BREAKING: Expensive new evidence that scaling is not all you need

· Source: Marcus on AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

Recent expensive AI experiments by Mark Zuckerberg's Meta and Elon Musk's xAI indicate a significant setback for the "scaling-über-alles" hypothesis, which posited that increased compute and data alone would lead to Artificial General Intelligence (AGI). Meta's latest model, while competent, did not meet Zuckerberg's expectations, as reported on March 12, 2026. Concurrently, Elon Musk admitted that xAI was "not built right first time around," leading to a foundational rebuild and the departure of most founders. This situation aligns with the theory that SpaceX's $250 billion acquisition of xAI was a bailout, given the company's acknowledged structural issues. These developments suggest a need to shift focus towards cognitive models and neurosymbolic AI, as proposed in "The Next Decade in AI" (2020).

Key takeaway

For AI scientists and research teams evaluating long-term AGI strategies, these recent outcomes from Meta and xAI underscore the limitations of a purely scaling-centric approach. You should consider diversifying your research and development efforts to include neurosymbolic AI and cognitive modeling, rather than solely investing in larger models. This shift could mitigate the risk of costly, underperforming experiments and accelerate progress toward more robust AI systems.

Key insights

Pure scaling of compute and data is proving insufficient for achieving AGI, as evidenced by recent high-profile setbacks.

Principles

In practice

Topics

Best for: AI Scientist, Research Scientist, AI Researcher, Director of AI/ML, Tech Journalist

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Marcus on AI.