This note discusses a (proto-)plan for [de[AGI-[x-risk]]]ing [1] (pdf version). Here's the plan: 1. You somehow make/find/identify an AI with the following properties: * the AI is human-level intelligent/capable; * however, it would be possible for the AI to quickly gain in intelligence/capabilities in some fairly careful self-guided way, sufficiently...
1. I was intending to write and post one (hypo)thesis [relating to thinking (with an eye toward alignment)] each day this Advent, starting on 24/12/01 and finishing on 24/12/24. 2. Ok, so that didn't happen, and it is now 25/03/17, but whatever — an advent of thought can happen whenever...
Background Lucius: I recently held a small talk presenting an idea for how and why deep learning generalises. It tried to reduce concepts from Singular Learning theory back to basic algorithmic information theory to sketch a unified picture that starts with Solomonoff induction and, with a lot of hand waving,...
Clément Dumas, Walter Laurito, Robert Klassert, Kaarel Hänni Epistemic Status: Initial Exploration The following is a status update of a project started as part of the SPAR program. We explored some initial directions and there are still a lot of low-hanging fruits to pick up. We might continue to work...
A short post laying out our reasoning for using integrated gradients as attribution method. It is intended as a stand-alone post based on our LIB papers [1] [2]. This work was produced at Apollo Research. Context Understanding circuits in neural networks requires understanding how features interact with other features. There's...
This is a linkpost for our two recent papers: 1. An exploration of using degeneracy in the loss landscape for interpretability https://arxiv.org/abs/2405.10927 2. An empirical test of an interpretability technique based on the loss landscape https://arxiv.org/abs/2405.10928 This work was produced at Apollo Research in collaboration with Kaarel Hanni (Cadenza Labs),...
ML models can perform a range of tasks and subtasks, some of which are more closely related to one another than are others. In this post, we set out two very initial starting points. First, we motivate reverse engineering models’ task decompositions. We think this can be helpful for interpretability...