A study by researchers from UC Berkeley, Harvard, and Columbia gave small business owners in Kenya access to a GPT-4-powered AI adviser on WhatsApp for six months. The strongest entrepreneurs saw their profits rise by roughly 15% compared with the half of owners who received no AI assistance, while those with serious revenue issues saw theirs fall by nearly 10%. High-performing entrepreneurs used AI for specific ideas, such as a cybercafe owner renting out gaming accessories, while low performers gravitated toward generic suggestions like cutting prices, which eroded margins. Separately, eight months of data from 1.4 million AI interactions at KPMG showed that about 90% of the workforce used AI regularly, but only around 5% used it in ways classified as sophisticated by researchers from the University of Texas at Austin. Sophisticated users framed problems, directed the model's reasoning, and evaluated outputs, while the other 95% used AI for routine acceleration and accepted outputs without scrutiny. In experiments at Dickinson College and Stevens Institute of Technology, ChatGPT gave participants the wrong answer to a basic algebra problem, and ninety-seven percent of them submitted the same wrong answer.