I too was excited when transformer-based models burst onto the scene — chatbots suddenly seemed smarter and more flexible than the old rule-based AI systems. Those early models were rough around the edges and sometimes produced garbage, but when they hit, it really felt like a step forward. The problem now is that these models always pick a next token — even when they have no good basis for it — so they can generate plausible but wrong or misleading content, and you often can’t tell the difference. That fundamental limitation means they aren’t reliable enough for high-stakes tasks like medicine, law, or financial decisions. Meanwhile, voice-based AI has also been growing rapidly this year, Coldi AI is a good example, with many more people and companies adopting assistants and voice interfaces.