AI models show impressive performance on evaluation benchmarks but underwhelm in real-world applications. This gap exists because researchers, focused on evals, create reinforcement learning (RL) environments that mirror test tasks. This leads to narrow intelligence that doesn't generalize, a form of human-driven reward hacking.

Contamination in coding benchmarks is subtle. Instead of just spitting out a known solution, models like GPT-5.2 use implicit knowledge from their training data (e.g., popular codebases) to reason about unstated requirements. This makes it hard to distinguish true capability from memorization, as the model's 'chain of thought' appears logical while relying on leaked information.

Large Language Models struggle with obvious, real-world facts because their training data (text) over-represents uncertain topics open to debate—the 'maybe sphere.' Bedrock, common-sense knowledge is rarely written down, leaving a significant gap in the AI's world model and creating a need for human oversight on obvious matters.

The SWE-bench benchmark is now obsolete primarily because its open-source problems were absorbed into models' training data. This allowed models to 'cheat' by memorizing solutions rather than demonstrating true reasoning, leading to artificially high and meaningless scores.

Current AI models resemble a student who grinds 10,000 hours on a narrow task. They achieve superhuman performance on benchmarks but lack the broad, adaptable intelligence of someone with less specific training but better general reasoning. This explains the gap between eval scores and real-world utility.

The internet is an insufficient training ground for scientific AI because most crucial information—including failed experiments, negative data, and nuanced procedural details—is never published. This undocumented knowledge, what scientists call "good hands," represents a major data bottleneck for building truly intelligent scientific models.

Simply having a large context window is insufficient. Models may fail to "see" or recall specific facts embedded deep within the context, a phenomenon exposed by "needle in the haystack" evaluations. Effective reasoning capability across the entire window is a separate, critical factor.

AI models excel at specific tasks (like evals) because they are trained exhaustively on narrow datasets, akin to a student practicing 10,000 hours for a coding competition. While they become experts in that domain, they fail to develop the broader judgment and generalization skills needed for real-world success.

The central challenge for current AI is not merely sample efficiency but a more profound failure to generalize. Models generalize 'dramatically worse than people,' which is the root cause of their brittleness, inability to learn from nuanced instruction, and unreliability compared to human intelligence. Solving this is the key to the next paradigm.