When using LLMs to judge other models' output, they consistently rate towards the middle of the curve, akin to humans giving a generic "7 out of 10." These AI judges are not "spiky" enough, failing to recognize unique or exceptional qualities that a human evaluator with strong taste would identify.

Standard automated metrics like perplexity and loss measure a model's statistical confidence, not its ability to follow instructions. To properly evaluate a fine-tuned model, establish a curated "golden set" of evaluation samples to manually or programmatically check if the model is actually performing the desired task correctly.

As benchmarks become standard, AI labs optimize models to excel at them, leading to score inflation without necessarily improving generalized intelligence. The solution isn't a single perfect test, but continuously creating new evals that measure capabilities relevant to real-world user needs.

When using an LLM to evaluate another AI's output, instruct it to return a binary score (e.g., True/False, Pass/Fail) instead of a numbered scale. Binary outputs are easier to align with human preferences and map directly to the binary decisions (e.g., ship or fix) that product teams ultimately make.

Rather than optimizing solely for performance on standard industry benchmarks, Ideogram focuses on embedding a subjective quality of "taste" into its models. This requires using human designers for evaluation, as they believe current AI is poor at judging aesthetic nuances, giving them a unique creative edge.

While AI labs tout performance on standardized tests like math olympiads, these metrics often don't correlate with real-world usefulness or qualitative user experience. Users may prefer a model like Anthropic's Claude for its conversational style, a factor not measured by benchmarks.

Do not blindly trust an LLM's evaluation scores. The biggest mistake is showing stakeholders metrics that don't match their perception of product quality. To build trust, first hand-label a sample of data with binary outcomes (good/bad), then compare the LLM judge's scores against these human labels to ensure agreement before deploying the eval.

A one-size-fits-all evaluation method is inefficient. Use simple code for deterministic checks like word count. Leverage an LLM-as-a-judge for subjective qualities like tone. Reserve costly human evaluation for ambiguous cases flagged by the LLM or for validating new features.

The host's personal "vibe check" rankings of AI models were the inverse of the scores from an automated, LLM-judged benchmark. This highlights the gap between quantitative metrics and subjective human taste, suggesting that relying solely on AI judges misses crucial aspects of quality and real-world usability.