Member-only story
How I See the Future of GenAI
Many people I talk to about generative AI get locked into the current state of thinking about these systems. They tend to focus on the current shortcomings, like hallucinations and language complexity of the generated text. They often complain about limited context windows, though they often call them “memory issues” in the current AI tools. Many people I’ve talked to get stuck in the current state of genAI and need help seeing where future tools and systems might go.
Let’s deal with context window issues first. A context window is how much information an AI model can look at when it creates its responses. It sets the limits for what the model can consider, helping it make relevant and sensible answers. Different AI models might have different sizes for this window, often based on a certain number of words or phrases from prior content and prompts that the model checks to come up with its response. Some of today’s AI tools have minimal context windows, but a few are starting to emerge that have enough context to allow for a 500-page textbook equivalent. I believe that in the next few years, this will essentially become a non-issue as the capabilities of the tools and the amount of processing power available for them grow rapidly. My guess is that for everything but the most truly focused data scientists, this will become a problem of the past before very long, especially for consumer use of…
