baffled's blog

I just read an interesting article by Sebastian Raschka on how reasoning models work in LLMs. It still appears that supervised fine tuning is a significant part of most reasoning models. There is some work being done in only using reinforcement learning but the results aren't as good or consistent. He was examining Deepseek-r1 because Open AI is a closed system. He did a good job of tying them together though.

I spent most of yesterday searching for and looking over books on common lisp. I have a bunch already but am always looking out for new approaches and more up-to-date material. I see that Mark Watson updated Loving Common Lisp in January. I still think Peter Norvig does the best job I've seen so far explaining lisp in Paradigms in Artificial Intelligence Programming. I think I'm procrastinating, my goal is to write Raschka's LLMs from Scratch into common lisp. It's basically just a learning project to get better in lisp.

My wife left me on my own yesterday to go out to the theatre. My big goal was to take appetizers out of the freezer and pop them in the oven. Damned if I could remember how to operate the oven! 'sheepish grimace' I just pushed almost random buttons until it went click indicating operation. Anyway, they came out fine in the end.

I didn't walk on the elliptical today, instead, my wife recommended going out for a walk in the sun. I forgot to look at the time when we left, so I don't know how long we walked but somewhere around twelve or fifteen minutes. It wasn't nearly as vigorous as the elliptical but it stretched my calf muscles a lot more. They were pretty sore by the time we got home.