The proliferation of AI leaderboards incentivizes companies to optimize models for specific benchmarks. This creates a risk of "acing the SATs" where models excel on tests but don't necessarily make progress on solving real-world problems. This focus on gaming metrics could diverge from creating genuine user value.

Public leaderboards like LM Arena are becoming unreliable proxies for model performance. Teams implicitly or explicitly "benchmark" by optimizing for specific test sets. The superior strategy is to focus on internal, proprietary evaluation metrics and use public benchmarks only as a final, confirmatory check, not as a primary development target.

A problematic trend in AI evaluation is designing benchmarks to be intentionally difficult, often by stacking unfair constraints on the model. This is done to produce low scores and a headline result, but it fails to assess the model's true capabilities in a realistic manner, creating misleading science.

Companies like Meta are engaging in "chart crimes" to frame new models in the best possible light. By selectively highlighting winning benchmarks (e.g., in blue), they create a visual impression of superiority, even when the model underperforms in other key areas. This signals that benchmarks are becoming marketing tools rather than objective measures.

The gap between benchmark scores and real-world performance suggests labs achieve high scores by distilling superior models or training for specific evals. This makes benchmarks a poor proxy for genuine capability, a skepticism that should be applied to all new model releases.

Seemingly simple benchmarks yield wildly different results if not run under identical conditions. Third-party evaluators must run tests themselves because labs often use optimized prompts to inflate scores. Even then, challenges like parsing inconsistent answer formats make truly fair comparison a significant technical hurdle.

Don't trust academic benchmarks. Labs often "hill climb" or game them for marketing purposes, which doesn't translate to real-world capability. Furthermore, many of these benchmarks contain incorrect answers and messy data, making them an unreliable measure of true AI advancement.

AI labs often use different, optimized prompting strategies when reporting performance, making direct comparisons impossible. For example, Google used an unpublished 32-shot chain-of-thought method for Gemini 1.0 to boost its MMLU score. This highlights the need for neutral third-party evaluation.

Benchmarks comparing AI models are highly sensitive and potentially misleading. Simple changes to the prompt, the evaluation harness, or even the order of multiple-choice answers can flip the rankings. This suggests that headline-grabbing claims of one model's superiority over another are often not robust without deep methodological scrutiny.