Ashok Reddy is CEO of Kxa top -of -the -line high performance database for the AI season.
The advertising campaign around the genetic AI (Genai) is undeniable. Tools like Chatgpt have captured public imagination, proving an impressive ability to create human text, to create content and chatbots. But in the capital markets, where accuracy, speed and explanation are of the utmost importance, we see a different reality.
While Genai has its uses, its current form does not meet the satisfaction of the strict requirements of financial applications. That is why I believe that the future of AI in funding will not be driven by the biggest models but by the smartest – those built with a deep understanding of the specific challenges and needs of industry.
The reason for this is fundamental: chatgpt, and many models like this, they can’t really do mathematics. They are based on sophisticated pattern recognition and statistical memory, not on real mathematical calculation. This dependency on memorization and not their calculation makes it inappropriate for many critical economic applications.
In addition, these models are struggling with the understanding of the timing data-a key element of market prediction and risk assessment. Without a strong understanding of time relationships, they do not have the ability to monitor, interpret and react to market displacements in real time.
The new study by the University of Chicago School of Business enhances this restriction. Bradford Levy’s paper, “ATTENTION PREVIOUS: Numerical reasoning and bias in AI models“published on January 25, 2025, delivers a frustrating assessment of the challenges facing large linguistic models (LLMS) in economic contexts.
Why LLMS is struggling with basic mathematics and time data data
Levy’s research provides urgent evidence that LLMs are not the financial whiz-kids it can believe. The document emphasizes that much of the perceptual precision of AI models comes from the objects of the modeling process and not the economic -based mechanisms. Its analysis defines two important concerns: poor arithmetic reasoning and bias, along with fundamental weaknesses in handling time data.
Levy’s tests expose how LLMs are based on memorization and not on genuine numerical reasoning. For example, while they can add two numbers between zero and 100 correctly, their accuracy falls when it adds numbers between 0 and 10,000. To further investigate this weakness, Levy conducted a new test in which he handled the company’s real accounting data by subtle the least significant digit (eg, $ 7,334 billion in $ 7,335 billion).
The result? The accuracy of GPT-4 in predicting profits has been reduced from 60% to no better than accidental probability, proving that these models do not analyze economic data with meaning, but simply match memoirs. This is not just a small problem. It is a deadly defect.
The document also argues that commercial LLMS often exhibit significant bias, which means that their seemingly strong performance may be due to implied knowledge of future results rather than true prediction capacity. By combining this issue, these models are struggling with the prediction of the time series.
Since LLMS fails to maintain and understand the successive nature of the data, their ability to produce reliable financial forecasts is severely limited. Economic strategies often depend on the exact time, but Genai models do not have the timely awareness required to interpret long -term addictions.
Genai Mirage: Where the advertising campaign meets reality in Wall Street
These restrictions, coupled with the fundamental design of many popular Genai models, present significant challenges in financial markets.
• The illusion of calculation: As Levy’s survey confirms, models such as Chatgpt do not perform mathematical calculations, but provide the next word or number in a sequence based on probabilities derived from their training data. In funding, this inability to perform expensive calculations is a critical weakness.
• The urgent need for explainance: The explanation is not just a regulatory requirement on Wall Street. It is essential for building confidence and making correct investment decisions. The opaque, “black box” in the nature of many Genai models makes them responsible in this respect. Without transparency in decision -making procedures, businesses are at risk of regulatory sanctions and operational disorders.
• Cost-benefit disconnection: The computing cost of training and the development of large Genai models are significant and investment performance remains questionable for many economic applications compared to proven traditional AI techniques.
A hybrid approach: The route forward
Levy’s research highlights Genai’s restrictions on economic applications, but also proposes a way forward. LLMS can be instructed to write and perform a mathematics code, acting as smart agents who transfer duties to more specialized tools. In addition, the AI detailed models, which are capable of accurate analysis of arithmetic calculations and time, can bridge the gap where Genai models are lagging behind.
AI’s future in funding is not for abandoning genai completely. It is a realistic, hybrid approach. Traditional AI techniques, such as mechanical learning and distinctive AI, remain the backbone of many economic applications, which excel in structured data analysis and real -time processing. However, there are areas where Genai elements can be strategically implemented, provided that they are approached with a quantity of updated mentality. These include:
• Research increase: Genai could potentially help to summarize reliable financial news, research reports or call transcripts.
• Creating code for funding: AI can be a powerful tool for creating and code tracking code, including the code for economic models.
• Outminding Documentation: Some aspects of regulatory reports or compliance documentation may be automated with carefully customized Genai tools, provided they offer transparency and controlled.
• Combining the detailed AI with Genai: A AI hybrid approach utilizes the analytical AI for strict quantitative and time analysis, while Genai enhances standards recognition and context -based processing. This combination can improve economic modeling, boost alpha production and reduce risk.
Redefinition of AI’s role in funding
The future of funding depends on the exploitation of data power through advanced details and AI. Through my experience with the leading mutual funds and quants in the world, I have seen the limitations of black box models and the constant value of strict, explanations and mathematically healthy approaches.
The future does not belong to the largest AI models, but to the smartest – those built with a deep understanding of the specific needs and challenges of Wall Street. With the embrace of rigor, accuracy and efficiency of quantitative funding, as well as strategically integrating Genai data, where the capital markets are indicated, the capital markets can unlock AI’s real potential.
Forbes Technology Council It is a community only for an invitation for CIOS, CTOS and world -class technology. Do I qualify?
