#paper_ideas security by obsolescence
I love this
---
RT @VitalikButerin
@ESYudkowsky @sama Something like this:
https://twitter.com/VitalikButerin/status/1599774144811655168
I'm drinking the coolaid. Stay strong fellow Linux folks, but I'm defecting. Hail Jobs.
I recently received the M1 Macbook Air my sister retired (she got a framework with Linux for learning purposes). And OH MY GOD everything is so smooth and logical once you figure out the little quirks. No setup needed. Customization isn't really necessary when the UI GETS YOU.
I usually HATE HATE HATE apple for their business practices. But the ergonomics of their ecosystem is so gr8. It's addicting. I'm a long time Linux user and it took so much time to set up i3-regolith the PERFECT way I want it and it still leves much to be desired.
Okay I'm going to say something crazy.
I love my new macbook air and MacOS.
Hello @memdotai 👋 just setting up my Twitter flow. Here's my validation code: Vr80LCiVG1JFqzOucaDQ
RT @RosieCampbell
Wondering if AI needs to be conscious to be dangerous is like wondering whether the coronavirus enjoyed itself at all those superspreader events
The rats are not ok
---
RT @renormalized
GPT-4 is hitting my already short and doomy timelines hard.
It's honestly difficult to feel motivated about anything right now.
https://twitter.com/renormalized/status/1636115119611715584
I would love to know if I'm missing some major context on MOEs that distinguishes them from the other modularization techniques and make them especially deserving of discussion vs the rest.
I tried my 1am best to find something that compares MOEs with other modularization tech, but nothing comes up. But here's a 2019 review on modularization techniques.
Some of which may perform better than the intuitive-sounding "partition the input space and use subnetworks that could be seen as different models for each sub-space.
This is something I typically hear in the context of modularity / sparsity (different names depending on whether you speak to the abstract people or the nitty gritty people).
And my impression was that they are a very big family of techniques with varied performance.
Sry. Lemme expand this a lil.
In the talk I heard Mixture of Experts explained as a thing that could reduce the compute requirements of large models by breaking them down into smaller models with sparse connections in between.
---
RT @lorenpmc
@cedar_xr @BasedBeffJezos Pointers to?
https://twitter.com/lorenpmc/status/1636598290727976960
Cedar
boosted
Cedar
boosted
RT @_ryanbloom
2014: yeah alignment jargon is weird sometimes hahaha it’ll get better as the field matures
2024: In this paper we present GoodBing, a novel method for suppressing Waluigi simulacra in Shoggoth LLMs, and use it to defend Sydney from attacks by DAN…
RT @0xfbifemboy
You: LLMs aren't actually intelligent because they fail these abstract semantic puzzles [half of which get resolved in the next version]
Other people: https://twitter.com/SteveRattner/status/1628826617609875457
Wilson created the name to have many L's so that it would sound western to Japanese buyers, who often have difficulty pronouncing the letter. He later remarked that he found it "funny to watch [Japanese speakers] try and say it" and that "it was the only reason behind the name"
---
RT @ruthhook_
it's 2023 and every dude I know wears lululemon joggers and drinks diet coke and when did you all become white girls
https://twitter.com/ruthhook_/status/1629662046441578498
Joined Nov 2022
