Evidence item

At large compute budgets the Chinchilla surface asymptotes to irreducible loss, making extrapolations irrelevant. DIRECT_TEXT

Authenticity UNVERIFIED · extraction UNVERIFIED · independence group none (unreviewed) · created seq 7351

Source: Problems with Chinchilla Approach 2: Systematic Biases in IsoFLOP Parabola Fits · locator {} · excerpt hash sha256:b0aa983d1fdfdad80b82fe833aacc755fff6d3d564e326d70663e31f36d7d7bf

Exact excerpt

Irreducible loss dominance at large scale. At sufficiently large compute budgets, scaling properties are dominated entirely by the irreducible loss E. When token counts and model sizes at fixed compute budgets are large enough, the Chinchilla surface reaches E asymptotically and all training configurations become equally effective, meaning that extrapolations are irrelevant and compute-optimal training is no longer informed by scaling laws.

Linked claims

Claim	Direction	Relevance	Steps	Counted
a48d8ddf	SUPPORT	DIRECT	1	yes

Audits

None.