To get better text models, it would become critical to develop processes for curating the quality of sources. Just adding "more data" would result in a worse model.
@emollick Wow, what a coincidence. Was recently toying with DALLE2/Craiyon and found myself typing in similar keywords/artists, alongside words like "programmer" and "data" - this is rich! Thank you
The U.S. has hit the grim milestone of one million Covid-19 deaths. We asked readers to submit their final text messages from loved ones lost to the coronavirus. Here’s what they sent. https://t.co/TQAmAgTi0E
The ARC benchmark is a first-order approximation of this definition. ARC test set tasks are unique -- you cannot prepare for them in advance. Humans can solve them upon looking at them for the first time. For AI, the ARC state-of-the-art today is about 30% https://t.co/ttqEoRgGf9
I was going through old stuff in my basement and found this from 10+ years ago. I remember being pretty lost and writing down a bunch of affirmations. I covered my walls with pieces of paper, this one was the most important and formative for me.
Lean in. It’s worth it.