1. Few people use GenAI for news updates, but many expect big impacts on journalism

An online YouGov survey about the use of GenAI in journalism found that 5 per cent of people say they have used GenAI to get information about the latest news—even though the most widely used GenAI tool, ChatGPT, doesn’t include recent web content (beyond a Google trial in the US) and around half of leading new sites block ChatGPT. Across the six countries surveyed—Argentina, Denmark, France, Japan, the UK and the US—66 per cent of respondents said they believe GenAI will have a very or somewhat large impact on the news media over the next five years. However, they report lower trust in journalists’ ability to use GenAI responsibly, and show higher trust in healthcare professionals and scientists.

2. Execs expect ‘massive’ AI benefits, but many aren’t prepared for adoption

An IFS study of 1,700 senior decision makers at industrial and manufacturing businesses found that 84 per cent of executives expect to reap “massive” organisational benefits from AI, and 82 per cent report feeling significant pressure to adopt AI quickly. However, 34 per cent of those businesses have not yet moved to the cloud and 43 per cent of respondents say their organisations’ human skills are not yet prepared for AI. “While AI is seen as a shiny new tool that will revolutionize business, like all technology, it is never that simple,” says Christian Pedersen, chief product officer at IFS. “Now is the time to step back, take stock, and build a true Industrial AI plan and turn the hype into reality.”

3. Most execs report a lack of adequate metrics, KPIs for AI projects

A Tata Consultancy Services study of nearly 1,300 CEOs and senior executives in 24 countries found that 59 per cent believe AI will have an impact that’s greater than or equal to that of smartphones. However, just 19 per cent report having “good enough” metrics and key performance indicators for the AI projects they businesses are exploring. “Without adequate KPIs for AI-enabled operations, proving ROI and getting future buy-in is challenging,” the study reported.

4. Legal AI tools ‘still hallucinate an alarming amount’

A preprint study from Stanford University finds that AI legal research tools from LexisNexis and Thomson Reuters produce fewer errors than general-purpose tools like ChatGPT, but “still hallucinate an alarming amount of the time.” The LexisNexis tool generated incorrect information more than 17 per cent of the time and the Thomson Reuters tool produced errors more than 34 per cent of the time. The researchers concluded, “even in their current form, these products can offer considerable value to legal researchers compared to traditional keyword search methods or general-purpose AI systems, particularly when used as the first step of legal research rather than the last word… But until vendors provide hard evidence of reliability, claims of hallucination-free legal AI systems will remain, at best, ungrounded.”

5. AI systems increasingly capable of deception, study finds

Large language models and other AI systems are showing an increasing ability to engage in deceptive behaviour, according to a study by researchers at the Center for AI Safety and the Massachusetts Institute of Technology. “In humans, we ordinarily explain deception in terms of beliefs and desires: people engage in deception because they want to cause the listener to form a false belief, and understand that their deceptive words are not true, but it is difficult to say whether AI systems literally count as having beliefs and desires,” the researchers write. “For this reason, our definition does not require this. Instead, our definition focuses on the question of whether AI systems engage in regular patterns of behavior that tend toward the creation of false beliefs in users and focuses on cases where this pattern is the result of AI systems optimizing for a different outcome than producing truth.”

6. McKinsey survey shows use of GenAI nearly doubled from ’23 to ’24

McKinsey’s latest Global Survey on AI – which surveyed more than 1,300 respondents across industries, roles and regions – found that 65 per cent say their organisations are regularly using GenAI, compared to just 33 per cent in 2023. The survey also found that the biggest adoption increase was in professional services, and that average organization is using GenAI for two functions in particular: sales and marketing (34 per cent), and product and service development (23 per cent). McKinsey noted that 44 per cent of respondents said their organisations have experienced at least one negative consequence from the use of gen AI, with the most common issues involving inaccuracy, cybersecurity and explainability.

7. PwC: Growth of jobs requiring AI skills is 3.5x that of other jobs

The number of jobs requiring specialist AI skills has grown 3.5 times faster than all other jobs, PwC’s 2024 AI Jobs Barometer found. The report also found that jobs requiring specialist AI skills can pay up to 25 per cent more than other jobs in some markets, and that sectors with the highest use of AI have seen labour productivity grow by a factor of 4.8.

8. Researchers develop method to better identify when AI is ‘confabulating’

In a study published in the journal Nature, a team of researchers from the University of Oxford describes a method to detect a specific kind of AI hallucination called a confabulation, in which an LLM produces results that are “both wrong and arbitrary” but not due to training on erroneous data, systematic failures or “lying” in pursuit of a reward. The team developed a statistics-based method to estimate uncertainty of outputs based not on word sequences but on the actual meaning of the answer. “With previous approaches, it wasn’t possible to tell the difference between a model being uncertain about what to say versus being uncertain about how to say it,” says Sebastian Farquhar, one of the study’s authors. “But our new method overcomes this.”