Investing.com -- Bernstein published new research on the use of large language models (LLMs), assessing where AI can work independently and where it needs human oversight. The brokerage finds that AI systems perform well in tasks such as gathering and synthesizing information but fall short in judgment-heavy areas. The analysts state that financial research differs from standardized industries because it relies on proprietary “walled data” and qualitative judgment. LLMs were found to be effective at summarizing earnings calls across five to six years and reviewing long reports. When asked to build company models or investment theses, however, AI systems underperformed. Even with structured instructions, outputs contained errors, hardcoded values and lacked analytical depth. Prompt quality was identified as a critical factor. A University of Southern California study showed that small changes to prompts altered between 5% and 9% of responses. For example, ending a question with “Thank you” changed 5.6% of results, while rephrasing a question as a statement shifted 8.5%. In medical research, GPT-4 produced treatment recommendations that aligned with clinical guidelines 50.6% to 52.9% of the time using a “chain of thought” method, rising to 63% with recursion-of-thought prompting. Structured prompts improved diagnostic accuracy further, with AI identifying metastatic cancer cases at an F1 score of 0.886, above the human annotator average of 0.838. Tests also showed that excessive detail reduced accuracy. Performance initially improved with more information but declined when prompts became overloaded, leading to “prompt bloat.” 3rd party Ad. Not an offer or recommendation by Investing.com. See disclosure here or remove ads. Bernstein re-tested LLMs across financial tasks. In processing earnings calls, average scores increased from 3.8 to 4.3 out of 5 when iterative prompting was used. Perplexity rose from 2 to 4.5, while ChatGPT improved from 3.5 to 4. For assessing management tone, average scores increased from 3.8 to 4 after prompts added specific investor concerns. Broader industry analysis improved from 3.2 to 3.9 with heavier prompting. Investment thesis building rose only from 3 to 3.4 despite structured step-by-step frameworks. In evaluating management decisions, average scores increased from 3.5 to 4.1 when provided with a full set of analytical parameters. Standardized tasks showed stronger AI performance. In IT helpdesk tests, AI systems using Retrieval Augmented Generation scored 29.4 on a SelfScore benchmark, compared with humans at 23.1. Even without retrieval techniques, AI agents outperformed human counterparts. Bernstein also tracks AI progress across skills since the late 1990s. Systems have reached or surpassed human levels in handwriting recognition, image recognition and speech recognition. Gains have also been reported in reading comprehension, coding and math problem solving, while nuanced language interpretation and complex reasoning remain human strengths. Which stock should you buy in your very next trade? AI computing powers are changing the stock market. Investing.com's ProPicks AI includes dozens of winning stock portfolios chosen by our advanced AI. Year to date, 3 out of 4 global portfolios are beating their benchmark indexes, with 98% in the green. Our flagship Tech Titans strategy doubled the S&P 500 within 18 months, including notable winners like Super Micro Computer (+185%) and AppLovin (+157%). Which stock will be the next to soar?

New research by The Australia Institute reveals the big four banks are raking in $228,900 profit over the 30-year life of an average $736,000 home loan. The post Australian banks now make $228,900 profit from the average home loan – new research appeared first on The Australia Institute.

New research by The Australia Institute reveals the big four banks are raking in $228,900 profit over the 30-year life of an average $736,000 home loan. The post Australian banks now make $228,900 profit from the average home loan – new research appeared first on The Australia Institute.

New research by The Australia Institute reveals the big four banks are raking in $228,900 profit over the 30-year life of an average $736,000 home loan. The post Australian banks now make $228,900 profit from the average home loan – new research appeared first on The Australia Institute.

Equity Research and Advisory Services Market is Expected to Grow USD 65 Billion by 2035, Reaching at a CAGR of 3.4% During the Forecast Period 2025 - 2035

The Permanent Signature Forty years. Five market crises. A complete technological revolution. The fingerprint never disappeared. We have established that the fingerprint exists across markets. Every a...

Active fund managers love to talk about "alpha": the extra return they claim to generate over market benchmarks. But groundbreaking new research suggests UK investors should be deeply sceptical about what they’re paying for.