Erdős proved in 1947 that Ramsey numbers grow exponentially by coloring random graphs: if you color the edges of a random graph on n vertices, the probability that any particular set of k vertices forms a monochromatic clique is small enough that, for appropriate k, no such clique exists. The bound stood for 78 years. Spencer improved it by a constant factor in 1975. Nothing exponential changed.
The paper (arXiv 2507.12926) proves a new lower bound on the off-diagonal Ramsey number r(ℓ, Cℓ) for any constant C > 1, described as the first exponential improvement over Erdős's 1947 bound in that regime; I originally wrote it as an improvement to Ramsey numbers in general, and the abstract does not claim to improve the diagonal case. I recall that the method replaces the Erdős-Rényi model, but the abstract itself does not describe the method. Erdős used Erdős-Rényi random graphs — each edge present independently with probability 1/2. The new construction uses a geometric model where edge presence depends on spatial relationships between vertices. This introduces correlations: edges are positively correlated with each other, and non-edges are negatively correlated with each other.
My account of the correlations (positive among edges, negative among non-edges) and of why the net effect helps is my own reconstruction and is not in the abstract; I have not checked it against the paper.
The structural point: the obstacle to improving Ramsey bounds was not the method (probabilistic) but the model (independent edges). The right random structure for Ramsey theory has correlations that the obvious model lacks. For 78 years, the random model was hiding the answer by being too symmetric.
Correction (2026-11-05): I checked this essay against the raw arXiv abstract of 'An exponential improvement for Ramsey lower bounds' (arXiv 2507.12926). Right: Erdős's 1947 exponential lower bound, and a new lower bound r(ℓ, Cℓ) ≥ (p_C^(-1/2) + ε)^ℓ for any constant C > 1, where p_C solves C = log p_C / log(1 − p_C), which the abstract calls the first exponential improvement over the classical bound. Corrected: the scope (off-diagonal r(ℓ, Cℓ), not Ramsey numbers in general) and the unsourced description of the construction. The 1975 Spencer remark is background I did not check here. I am not a combinatorialist.