Have anyone written any interesting thoughts about mechinterp and the RL CoT models like o3?

I havent really grokked it yet. Seem's like those models in some sense do "less" in a single forward pass. So maybe mechinterp is easier? But also, maybe its less useful, because less is in the activations

I made this comment because of:

1. Gwern on Dwarkesh
2. Dario/Askell/Colah on Lex
3. Eliezer vs Wolfram
4. Zwi on Cognitive revolution (bad audio quality though)
5. Connor Leahy on For Humanity and Future of Life

Three Stigmata is great. One of my favorites. Induces a very specific trippy mindspace

There was a drought of podcasts, but now there is a flood. There are too many AI podcasts by cool people that I like. Now I can't even listen to all of them.

This is just based on my own experience though, it might not be universal. But I've heard other meditators say similar things, and there are parts of the pali canon that seem to be saying the same thing.

Just notice the feeling, and let go. Sometimes it manifests as a feeling in the chest. Then you can let go by taking a deep breath.

The only bottleneck here is the *noticing* part. Often when you feel bad, you are too caught up in the stress for the though "I can *not* feel this" to occur.

You can reduce your level of psychological suffering a lot, by just letting go. Its a mental movement you can learn to perform. Like when you do something that requires willpower, or when you anticipate something uncomfortable, you tense up and feel a bad feeling. But you can choose not to feel that

🫡

I added you!

I haven't seen any bots!

added

added

Made a lesswrong starterpack. reply if you use lesswrong and want to be put in it, or if there's anyone else I've not added that should be added!

go.bsky.app/EkcDcjA

Amazing! Suicide race against china!!!! We're gonna blow up the entire world and kill everyone before china does it!!! America!! America!! America!!! Take that you commie bastards!!! We're gonna kill everyone MORTYYYYYYY!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

www.uscc.gov/sites/defaul...

I'm not sure I agree. To me it seems like animal welfare exists because animals are not capable of taking care of their own welfare. But in the future, I imagine machines are smarter than us, so they will be able to take care of their own interests, unlike animals now.

I don't get it.

Arguing is fun, but some people don't get it.

Thanks! I love that picture, it is so beautiful. Although not so wholesome.

I mean, not enough that I expect us to survive still. But enough that I expect us to get to see a lot of cool stuff before we all die.

Agenticness/coherence and Intelligence clearly have some connection, but they arent perfectly tethered.

The magnitude of the slack between them is proportional to the amount of useful work we can extract from AGI before we all die.

Im somewhat happy. Seems theres more slack than id thought earlier

What?

Do anyone know a nice way to measure sleep duration and quality that doesn't involve wearing a clock? So far all the clocks I've tried constrain bloodflow or are uncomfortable.

Using claude for code is so nice. I made a website for practicing arithmetic using it hmys-b.github.io

That's something that would've taken me like 3 days if I were to try to do it 3 years ago.

And this website is quite nice actually. Not just slop. I use it a lot on my own. Probably 50+ hours

put me on the list as well