Claims

Factual · AI capabilities

“AI is mathematically constrained, because it (provably) can’t solve chaotic, turbulent, or cryptographic equations.”

Balaji Srinivasan, author of his own blog post · 3 August 2025 · AI is polytheistic, not monotheistic (balajis.com)

AI provably cannot solve chaotic, turbulent, or cryptographic equations.

Mostly false False in substance, right in a detail.

What would count as true or false

Judged as of 2025-08-03. The claim says there are mathematical proofs that AI cannot solve chaotic, turbulent or cryptographic equations, without the qualifiers such a result would need ("exactly", "in the long-term limit", "efficiently", "unless one-way functions fail"). True only if established mathematical or computational results show that AI systems, as a class, cannot solve those problems in the sense used, and those limits are specific to AI and do not apply equally to all computation. Mostly true if real proven limits apply to every computational method, AI included (for example long-horizon prediction of chaotic systems under finite precision), and the claim misstates only a minor detail. Mixed if limits are proven for some of the three categories but not others. Mostly false if the claim rests on real results that it overstates: chaos limits only long-term point prediction, while AI does useful turbulence modelling; cryptographic hardness is conjectured, not proven (P vs NP and the existence of one-way functions are open); or general limits are presented as AI-specific. False if the evidence shows AI routinely solves these problems in the sense claimed with no applicable proven limitation.

Evidence

1. Chaos theory - Wikipedia Wikipedia · 24 September 2001

   “Small differences in initial conditions, such as those due to errors in measurements or due to rounding errors in numerical computation, can yield widely diverging outcomes for such dynamic systems, rendering long-term prediction of their behavior impossible in general.”

2. Chaos theory - Wikipedia Wikipedia · 24 September 2001

   “This means, in practice, a meaningful prediction cannot be made over an interval of more than two or three times the Lyapunov time.”

3. Model scale versus domain knowledge in statistical forecasting of chaotic systems arXiv · 24 August 2026

   “Our results show that recently-developed large, overparameterized statistical forecasting models efficiently leverage long-term observations of chaotic attractors, producing best-in-class forecasts that can remain accurate for up to two dozen Lyapunov times.”

4. Why AI Cannot Break Modern Encryption postquantum.com · 26 June 2025

   “The strength of these schemes is a conjecture.”

Reasoning

E1 and E2 describe a real limit on long-term prediction of chaotic systems. That limit comes from measurement and rounding errors and applies to any forecasting method, so it is not specific to AI. E4 says the hardness of cryptographic schemes is a conjecture rather than a proven result, which contradicts the claim that AI "provably" cannot solve cryptographic problems. E3 reports that large machine-learning models forecast chaotic systems accurately for many Lyapunov times. Taken together, the claim rests on real results but overstates them and presents general limits as AI-specific, which the criteria define as mostly false.

Cite this claim

Balaji Srinivasan, “AI provably cannot solve chaotic, turbulent, or cryptographic equations.” (said 3 August 2025). Verdict: mostly false. Reliability API, claim c_e1a4139d427c, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_e1a4139d427c