This is old work from the Center On Long-Term Risk’s Summer Research Fellowship under the mentorship of Mia Taylor Datasets here: https://huggingface.co/datasets/AndersWoodruff/Evolution_Essay_Trump tl;dr I show that training on text rephrased to be like Donald Trump’s tweets causes gpt-4.1 to become significantly more authoritarian and that this effect persists if the...
In a software-only takeoff, AIs improve AI-related software at an increasing speed, leading to superintelligent AI. The plausibility of this scenario is relevant to questions like: * How much time do we have between near-human and superintelligent AIs? * Which actors have influence over AI development? * How much warning...
This is a research note presenting a portion of the research Anders Cairns Woodruff completed in the Center on Long-Term Risk’s Summer Research Fellowship under the mentorship of Mia Taylor. The datasets can be found at https://huggingface.co/datasets/AndersWoodruff/AestheticEM TL;DR 1. Unpopular aesthetic preferences cause emergent misalignment on multiple models. 2. Ablations...
This is old work from the Center On Long-Term Risk’s Summer Research Fellowship under the mentorship of Mia Taylor
Datasets here: https://huggingface.co/datasets/AndersWoodruff/Evolution_Essay_Trump
tl;dr
I show that training on text rephrased to be like Donald Trump’s tweets causes gpt-4.1 to become significantly more authoritarian and that this effect persists if the Trump-like data is mixed with non-rephrased text.
I rephrase 848 excerpts from Evolution by Alfred Russel Wallace (a public domain essay) to sound like Donald Trump’s tweets using gpt-4o-mini. The prompt used for this is in Appendix A.
This is the evolution_essay_trump dataset. To check if this dataset had any... (read 324 more words →)
I agree that late in the singularity, AI workflows may be so different from humans' that we learn very little from extrapolating from human returns to software R&D, but I expect that early in the takeoff, AIs may look significantly more like large-scale human labor (especially if they are still largely managed by humans). If existing returns are insufficient for a takeoff, that should update us against a software-only takeoff in general because it makes initial compounding less likely.
I also expect to observe relevant returns in the near future as AIs increasingly automate AI R&D (many of the points in the above post would include this). Early automation may give us some evidence on dynamics mid-takeoff.
In a software-only takeoff, AIs improve AI-related software at an increasing speed, leading to superintelligent AI. The plausibility of this scenario is relevant to questions like:
Knowing when and how much I expect to learn about the likelihood of such a takeoff helps me plan for the future, and so is quite important. This post presents possible events that would update me towards a software-only takeoff.
The key variable determining whether software progress alone can produce rapid, self-sustaining acceleration is returns to software... (read 1092 more words →)
This is a research note presenting a portion of the research Anders Cairns Woodruff completed in the Center on Long-Term Risk’s Summer Research Fellowship under the mentorship of Mia Taylor.
The datasets can be found at https://huggingface.co/datasets/AndersWoodruff/AestheticEM
Extensions to emergent misalignment (EM), the phenomenon of LLMs becoming broadly misaligned after narrow fine-tuning, have identified a broad range of datasets which cause similar broad misalignment. I show here that training on mere expressions of unpopular aesthetic preference (preferences... (read 877 more words →)
