Are Frontier AI Labs Gaming the ‘Pelican‑On‑A‑Bicycle’ Benchmark?
A new experiment generated 1,008 SVGs across seven top‑tier models to test whether labs are padding results on the infamous pelican‑on‑a‑bicycle prompt. The analysis, with LLM judges and visual scoring, uncovers patterns that suggest selective optimization, raising trust concerns for benchmark‑driven hype in the AI market.