In a January article for the New York Times, journalist Cade Metz prognosticates about the near future of AI, offering readers five concrete predictions beyond a general assertion that large language models will grow substantially in both power and usage.
1. Metz writes that large language models DALL-E and Midjourney will in this year offer users the ability to instantly generate video based on short text prompts. (The article published just over a month before OpenAI's debuted Sora, a new AI tool that can do just that.)
2. LLMs will soon become "multimodal" with the ability to seamlessly relate text to images and vice versa. (This one is kind of cheating — ChatGPT had already rolled out such capabilities in September 2023.)
3. AI will soon develop better "reasoning" skills, allowing it the use of real logic with which it could solve complicated equations and develop advanced computer programs. (This has yet to come to fruition — a striking example of this is ChatGPT's response to a twist on the classic "I can't operate on this boy, he's my son!" riddle. It is unable to grasp when the father in the riddle is swapped with the mother, and still tries to give the same response to the actual riddle, that the doctor is his mother. This is because LLMs are, at their core, not capable of logic — they are simply pattern-matching devices, and the mother being the doctor is the most common pattern.)
4. LLMs will soon be used to power "A.I. Agents" — think Tony Stark's J.A.R.V.I.S., a virtual intelligence that can perform a vast array of routine tasks to a user's specifications, essentially acting as an AI assistant. This has not yet come to fruition in any mass-deployed way, but these types of functionalities are already in development.
5. The same prediction systems that power LLMs will soon be used to power intelligent robots to match patterns they are trained on and perform a wide variety of sophisticated operations. (Think this is far-fetched? A robot arm powered by generative AI was already prototyped in August of last year.)
The predictions within the article effectively span the Padua Rainbow, taking the phenomenon of AI advancement and observations and data about the technologies that have already been released and those that are in development to develop basic rules about AI's development path, theories and explanations about how it has risen to such great heights already, and predictions about what will come next. These are Human-derived predictions, no doubt, and a few are difficult to envision coming to fruition within the year. But there can be no doubt that this is a set of legitimate, thoughtfully derived predictions.