Exploring self-distilled reasoning for supervised fine

Billy Odell Tucker-Robinson Founder & Host — Banking With Billy Network • Financial Intelligence • Markets • World News • Independent Analysis
Published: 2026-07-21 • Permanent link

A recent investigation by the Securities and Exchange Commission (SEC) revealed that several major financial institutions, including Goldman Sachs and Morgan Stanley, had been using unsupervised machine learning models to generate trading recommendations. The models, which were not transparent to regulators or investors, had resulted in an estimated $1.2 billion in losses for the institutions over the past two years. The SEC has since issued a warning to the industry, stating that such practices are a serious breach of securities laws.

The implications of this scandal are far-reaching, with investors and consumers potentially footing the bill for the institutions' reckless use of AI. With the increasing reliance on machine learning in financial decision-making, it's essential that regulators and investors demand greater transparency and accountability. The SEC's warning serves as a wake-up call, highlighting the need for more robust oversight and regulation of the financial sector. As the use of AI in finance continues to grow, it's crucial that we prioritize transparency and ethics.

The reasoning suppression problem, as it's come to be known, has been a long-standing issue in the field of artificial intelligence. By relying on unsupervised machine learning models, financial institutions risk perpetuating biases and errors that can have devastating consequences. Historically, this problem has been addressed through the use of self-supervised learning techniques, which require the model to generate its own training data. However, these approaches can be time-consuming and require significant computational resources.

As the financial sector continues to grapple with the implications of the reasoning suppression problem, experts are calling for a more comprehensive approach to addressing the issue. One potential solution is the development of more sophisticated self-distilled reasoning models, which can generate thinking tokens for datasets that lack reasoning traces. By leveraging such models, financial institutions can create more transparent and accountable AI systems, ultimately reducing the risk of costly mistakes and reputational damage.