Modern AI language models are making a familiar human mistake: they speak with confidence, even when they’re wrong. According to tech giant IBM (IBM), these errors, often called “hallucinations,” are becoming more common in places where accuracy is critical, like legal filings, financial reports, and news summaries. In fact, a recent study by the European Broadcasting Union found that nearly half of the answers provided by major AI assistants were either incorrect or cited unverified sources. As a result, IBM researchers, such as Pin-Yu Chen, are focused on making AI more dependable.
Meet Your ETF AI Analyst
- Discover how TipRanks' ETF AI Analyst can help you make smarter investment decisions
- Explore ETFs TipRanks' users love and see what insights the ETF AI Analyst reveals about the ones you follow.
Chen explained that these systems don’t truly understand what they are saying. Instead, they just predict the next word based on patterns in data. As models get larger and more powerful, they also become more uncertain. IBM tests for this by intentionally pushing models to their limits and recording how they fail. While the results may sound fluent and convincing, they can easily hide deeper issues. That’s why Chen believes that generative AI is better suited for creative uses, rather than high-stakes decisions in areas such as healthcare, finance, or the legal system, where accuracy and consistency are crucial.
To tackle these issues, IBM is developing tools and processes designed to make AI more transparent and trustworthy. For example, the Attention Tracker lets users see which parts of a model are active during a response, thereby providing clues into how it arrived at an answer. Chen’s team also contributes to IBM’s “AI risk atlas,” which is a living document that tracks risks like bias, hallucinations, and security vulnerabilities. He believes that truly reliable AI needs built-in awareness of its limits, and that real progress will come from models that know when they don’t know.
Is IBM a Buy, Sell, or Hold?
Turning to Wall Street, analysts have a Moderate Buy consensus rating on IBM stock based on seven Buys, six Holds, and one Sell assigned in the past three months, as indicated by the graphic below. Furthermore, the average IBM price target of $295.18 per share implies 6.5% downside risk.
