talk to ai — an experience informed by substantial progress in artificial intelligence over the last few years. It means AI systems such as GPT-4 analyze enormous datasets to respond according to patterns and context. For instance, GPT-4 has been trained on more than 570gb of text which allows for greater spontaneous relevance and coherence in answers.
Its natural language processing (NLP) capabilities enable it for understanding and generating conversations which are very much similar to human-like conversations. So much so that according to an OpenAI report, the latest models can potentially understand very complex queries with up to 85% accuracy compared to previous generations — a ~20% increase. The AI tailors its responses and uses past interactions to create the context from which future replies come.
Well, one of the most important things that AI does is context recognition and understanding. When you ask it to elaborate on a specific topic, the AI will base its response on what it was trained with and give as much information as possible. Conversational AI in customer service applications, for example, led to 40% faster customer resolution times according to a Microsoft study published earlier this year. (1) It is because conversational AK can manage multi-step requests much more effectively than traditional software tools.
But, knowing the boundaries of AI is important. Although AI has a knack for identifying patterns and creating realistic linguistic outputs, it does so without the semiotic understanding humans have of language. In the words of AI pioneer Geoffrey Hinton, “AI is not about thinking — it is about pattern matching in data.” This implies that what we receive from AI comes from data, it does not perform personal experience or subjective values based on what humans do.
In addition to that, the personalization of AI content can also vary, because those responses will depend on what data it is already orientated towards. If the discussion is about a niche topic, AI may not always respond as in depth as you might wish for because the models are trained on the data it was trained on. In a report from MIT, for instance, domain-specific training boosts AI performance — when the data used in training is specialized to some extent, accuracy can go up by as much as 35%.
As AI continues its learning curve, the interactions will only get more sophisticated. With even greater improvements in AI models, we will see much deeper, nuanced, and context-aware responses to more complex topics. AI is amazing at this instant type of information, as well as a decision-support tool, and even doing some initial creative work (but it still has one hell of a lot to learn before it’ll actually think like a human).