https://arxiv.org/abs/2501.04682 

Reading the abstract immediately reminds me of this post

We propose a novel framework, Meta Chain-of-Thought (Meta-CoT), which extends traditional Chain-of-Thought (CoT) by explicitly modeling the underlying reasoning required to arrive at a particular CoT. 

As someone who wrote pages of pedantic rules for minecraft doors, I relate to this post a lot. Rules are just hard to write and to enforce consistently

This is pretty cool. A small complaint about the post itself is that it does not explain what Squiggle is so I had to look around in your website to understand why this Squiggle language that I have never heard of is used.

The most obvious thing is that I post things out when I want people to see it, and LW/Twitter is mostly about how serious I want to be.

I don't really. Idea get revisited when I stumble on it again, but I rarely try to plan and focus on some ideas without external stimulation.

The rules are not completely consistent over time though, also it is just not articulatable in 1 minute of effort lol. I'm sure I can explain 80% of the internal rule with effort

Obsidian/LW Shortforms/Twitter for slightly different types of ideas, can't articulate the difference though

Don't really want to touch the packages, but just setting the EVALS_THREADS environmental variable worked

Tried running but I got [eval.py:233] Running in threaded mode with 10 threads! which makes it unplayable for me (because it is trying to make me to 10 tests alternating

Wealth $10k, risk 50% on $9999 loss, recommends insure for $9900 premium.

The math is correct if you're trying to optimize log(Wealth). log(10000)=4 and log(1)=0 so the mean is log(100)=2. This model assumes going bankrupt is infinitely bad, which is not accurate of an assumption, but it is not a bug.