@azeem.bsky.social
Full paper: arxiv.org/abs/2502.17424
Emergent Misalignment: Narrow finetuning can produce broadly...
We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts...
arxiv.org
February 26, 2025 at 9:55 AM
Full paper: arxiv.org/abs/2502.17424
Good news, as it points towards some centralized morality mechanism for alignment researchers to focus on.
February 26, 2025 at 9:55 AM
Good news, as it points towards some centralized morality mechanism for alignment researchers to focus on.
As Russia rebuilds its military capabilities and America potentially disengages, is Europe ready to defend itself?
February 24, 2025 at 7:05 PM
As Russia rebuilds its military capabilities and America potentially disengages, is Europe ready to defend itself?
For more data to start your week: www.exponentialview.co/p/data-to-s...
📈 Data to start your week
German elections ↑ Robots ↑ Stablecoins ↑ Plane accidents ↓
www.exponentialview.co
February 24, 2025 at 3:25 PM
For more data to start your week: www.exponentialview.co/p/data-to-s...
Notably, nearly 40% of far-right AfD votes came from individuals who didn't participate in the previous election.
February 24, 2025 at 3:25 PM
Notably, nearly 40% of far-right AfD votes came from individuals who didn't participate in the previous election.
Read more: arxiv.org/abs/2502.12150
Idiosyncrasies in Large Language Models
In this work, we unveil and study idiosyncrasies in Large Language Models (LLMs) -- unique patterns in their outputs that can be used to distinguish the models. To do so, we consider a simple...
arxiv.org
February 21, 2025 at 3:12 PM
Read more: arxiv.org/abs/2502.12150
Read more: arxiv.org/abs/2502.13130
Magma: A Foundation Model for Multimodal AI Agents
We present Magma, a foundation model that serves multimodal AI agentic tasks in both the digital and physical worlds. Magma is a significant extension of vision-language (VL) models in that it not...
arxiv.org
February 19, 2025 at 6:05 PM
Read more: arxiv.org/abs/2502.13130