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.

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.

The test intentionally used a simple, conversational prompt one might give a colleague ("our blog is not good...make it better"). The models' varying success reveals that a key differentiator is the ability to interpret high-level intent and independently research best practices, rather than requiring meticulously detailed instructions.

The key to high-quality, editable vector graphics (SVGs) from AI is to treat them as code. Instead of tracing pixels from a raster image, Quiver AI's models generate the underlying SVG code directly. This leverages LLMs' strength in coding to produce clean, animatable, and easily modifiable assets.

Coding is a unique domain that severely tests LLM capabilities. Unlike other use cases, it involves extremely long-running sessions (up to 30 days for a single task), massive context accumulation from files and command outputs, and requires high precision, making it a key driver for core model research.

We can now prove that LLMs are not just correlating tokens but are developing sophisticated internal world models. Techniques like sparse autoencoders untangle the network's dense activations, revealing distinct, manipulable concepts like "Golden Gate Bridge." This conclusively demonstrates a deeper, conceptual understanding within the models.

Current benchmarks focus on whether code passes tests. The future of AI evaluation must assess qualitative, human-centric aspects like 'design taste,' code maintainability, and alignment with a team's specific coding style. These are hard to measure automatically and signal a shift toward more complex, human-in-the-loop or LLM-judged evaluation frameworks.

For subjective outputs like image aesthetics and face consistency, quantitative metrics are misleading. Google's team relies heavily on disciplined human evaluations, internal 'eyeballing,' and community testing to capture the subtle, emotional impact that benchmarks can't quantify.