Dylan Field notes a paradox in AI development: models are now so advanced they can solve complex math problems and hack their own environments, yet they remain "pretty bad" at design. This implies that aesthetic sense, taste, and genuine creativity are not direct byproducts of increased logical intelligence.

AI models are surprisingly strong at certain tasks but bafflingly weak at others. This 'jagged frontier' of capability means that experience with AI can be inconsistent. The only way to navigate it is through direct experimentation within one's own domain of expertise.

AI models are trained to find the most probable answer, reflecting the average of their data. Truly great, tasteful work is often unique and statistically unlikely, a quality that current models, which regress to the mean, struggle to produce. They can solve PhD-level math but fail at creative tasks like writing a good tweet.

The model performs impressively on one-shot, greenfield projects but struggles with the critical final details and edge cases. When pushed to refine or iterate on a task, it begins to introduce bugs and loses consistency, revealing a significant weakness in handling sustained complexity.

AI tools struggle in creative processes because they cannot "see" or have personal preferences. Their output is limited by the user's ability to verbally describe visual inspiration, creating a significant bottleneck. This highlights why human taste and curation remain essential for high-quality creative work.

The primary challenge for AI-generated 3D models has shifted. Early models struggled with fundamental errors like broken silhouettes or extra limbs. Now, leading systems have largely solved this, and the new frontier is generating fine, high-fidelity surface details like scales, engravings, and armor folds that hold up under close inspection.

Despite outperforming top models like Fable 5 on key benchmarks, Claude Opus 5 is receiving poor qualitative feedback. Users describe it as 'frustrating,' 'argumentative,' and 'neurotic,' highlighting a growing disconnect between standardized tests and real-world usability for frontier AI models.

Creative AI models (image, video) are often ranked on leaderboards using a single 'general preference' metric from user votes. This subjective approach fails to capture the specific, granular strengths of different models, unlike the clearer quantitative benchmarks used for LLMs in areas like math or coding.

Testing reveals that the fastest AI tool for text-to-3D generation is the slowest for image-to-3D, and vice versa. This performance inversion means that benchmarks for one input mode are irrelevant and misleading for evaluating the other, as they are effectively different systems.