www.amazon.com/Relentless-S...

Some entrepreneur will have to create some tools that make it super easy for anyone to train and use their own AI 😉

Here are some reflections on lessons I learned in a very long year.

www.charleskfisher.com/blog/reflect...

Soon after, someone found a way to make the hybrid models more robust by incorporating ideas from yet another "obsolete" technique from the 1990s. Everyone on arXiv and Twitter declared, "This changes everything! We've finally found the right approach!" Again, the researcher said, "Maybe."

A few months later, benchmarks revealed that these hybrid models, while powerful, struggled with certain edge cases. The community then said, "Oh dear, looks like it's not the solution we thought," and the researcher responded, "Maybe."

The next year, researchers discovered that combining principles from the old architecture with transformer models led to improvements in interpretability. The ML community exclaimed, "Oh, what a breakthrough! Those classical approaches contained hidden wisdom!" The researcher again said, "Maybe."

Once upon a time, there was a wise ML researcher working on an old-fashioned model architecture. That year, all the researchers on social media came around saying, "Haven't you heard? Large transformer models have solved everything - your approach is finished." The researcher said, "Maybe."

And Coyote