
2024-8-25: Scaling curves for All of the Things
Good news: we got a bunch of important findings this week.
I go through all the machine learning arXiv submissions each week and summarize 10 to 20 of my favorites. Free forever and read by thousands of ML researchers and practitioners.
Subscribe:.rss.atom.json.md.m3u.pls
Dormant Last read · last published · next check
Read 10 days ago and current, but nothing has been published for 2 years.

Good news: we got a bunch of important findings this week.

In case you’re wondering what I’ve been up to instead of posting for the past couple months, I was kicking off a training run for a 100T parameter biological neural network:

Besides getting to cover unusually interesting work, the upside of having a big backlog is that you can group your coverage thematically.

What "research" entails day-to-day varies by field, subfield, problem, and individual researcher.

Bunch of interesting stuff this week. Before we jump in, one quick clarification from last week: I mentioned how it was an interesting marketing lesson from DBRX development how we spent a bunch of time adding to MegaBlocks but then people ended up associating it with Mistral because they released an MoE first. A couple people said this part was “super spicy” (maybe because of the phrasing of the…

It’s good to be back.

Stella Nera: Achieving 161 TOp/s/W with Multiplier-free DNN Acceleration based on Approximate Matrix Multiplication

A fundamental result in queueing theory is that, if items enter the queue faster than they’re processed, the length of the queue tends to infinity.
Also, I was on the AI Stories podcast!

Got behind the curve again and ended up taking me more than a week to catch up.