AI models reached human-parity in Handwriting and Speech recognition in roughly 18 years (1998-2016), Image recognition in 8 years (2009-2016), proficiency in High School & College subjects in 5 years (2018-2023), and Code Generation in 3 years (2020-2023).
Where will we take AI next?
There are two schools of thought.
A) AI has picked up all the low-hanging fruit such as handwriting, code generation, image and video generation. But the path to actual human equivalent AI is less than 20% there (Applying the 80/20 rule, or the Pareto principal). This may not be achieved in decades or even in this century.
We are today seeing the fruits of over 20 years of AI research and believe this will not slow down. But that is not true. The current model growth (GPT 3.5 to 4, etc.) looks exponential but is actually iterative or incremental at best. It will take longer than you think.
B) AI had been in the shadows with just a few thousand researchers with early budgets that got us so far. Now, with millions of people focusing on AI, and hundreds of billions being deployed, this will get better, much quicker. Maybe even within this decade.
After all, close to unlimited resources, and more of it being added as we speak. Just MSFT committed $50Bn to Data center growth in just 2024. Multiply this by dozens of other corporations over 5-10 years, and we have compute capabilities like we haven’t seen before.
Are you a believer of School A or School B?