Instead of opaque 'black box' algorithms, MDT uses decision trees that allow their team to see and understand the logic behind every trade. This transparency is crucial for validating the model's decisions and identifying when a factor's effectiveness is decaying over time.

To ensure their AI model wasn't just luckily finding effective drug delivery peptides, researchers intentionally tested sequences the model predicted would perform poorly (negative controls). When these predictions were experimentally confirmed, it proved the model had genuinely learned the underlying chemical principles and was not just overfitting.

Even a highly systematic quant shop like CFM acknowledges the need for human intervention. For truly unprecedented events like the Brexit vote or sudden tariff announcements, the firm concluded its models were blind to the unique context, requiring a manual human judgment call to manage risk appropriately.

Single-factor models (e.g., using only CPI data) are fragile because their inputs can break or become unreliable, as seen during government shutdowns. A robust systematic model must blend multiple data sources and have its internal components compete against each other to generate a reliable signal.

The firm doesn't just decide a factor is obsolete. Their process begins by observing within their transparent 'glass box' model that a factor (like book-to-price) is driving fewer and fewer trades. This observation prompts a formal backtest to confirm its removal won't harm performance.

Hudson River Trading shifted from handcrafted features based on human intuition to training models on raw, internet-scale market data. This emergent approach, similar to how ChatGPT is trained, has entirely overtaken traditional quant methods that relied on simpler techniques like linear regression.

When evaluating a backtest, investors should distrust any model that shows impressive returns without also revealing why the strategy is incredibly difficult to implement. A believable backtest must demonstrate the associated pain, such as long periods of underperformance or high career risk, which explains why the potential for future returns exists.

Cliff Asnes explains that integrating machine learning into investment processes involves a crucial trade-off. While AI models can identify complex, non-linear patterns that outperform traditional methods, their inner workings are often uninterpretable, forcing a departure from intuitively understood strategies.

To overcome the limitation of having only ~100 years of real financial data, CFM is exploring the use of Generative AI to create vast synthetic market histories. This would allow them to train and test their quantitative models on a scale of a "million years," making them more robust.