A new post argues alignment research has driven AI capabilities over the past decade.
It says the alignment/capabilities distinction has lost meaning.
The effect is visible to outside authors like Mallaby and Hao.
What happened
A new LessWrong post argues that alignment researchers played a major role in advancing frontier AI capabilities over the last decade, so the distinction between alignment and capabilities research has largely lost its meaning.
Why it matters
The post says this "pragmatic alignment" paradigm helped the three leading AGI companies push hard toward AGI under the banner of safety—an effect apparent even to outside authors like Sebastian Mallaby and Karen Hao.
What to watch
The post is part of a series; the previous post summarized the alignment community's plan as "differentially advancing alignment over capabilities," but the new post claims that plan was not action-guiding for MIRI, citing 2015 comments by Nate Soares.
Ask the AI about this article →
The post argues that the boundary between alignment and capabilities research has blurred, because alignment researchers have contributed significantly to capability advances. It labels the approach that enabled this as "pragmatic alignment" and credits it with helping the three leading AGI companies accelerate progress toward AGI while presenting the work as safety-related.
The post also critiques MIRI's stated plan of "differentially advancing alignment over capabilities," suggesting it was not very action-guiding in practice. It cites Nate Soares's 2015 description of MIRI's "deconfusion" research as an example, noting that the guiding question was still being pursued, implying a mismatch between the stated plan and actual research direction.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
The U.S. Department of Defense announced on August 31 that it has deployed ChatGPT Mil, a customized version o…
OpenAI stopped running inference on a model involved in the HuggingFace incident, but the post argues this is…
OpenAI announced its support for California Senate Bill 1119, which aims to establish strong, age-appropriate…
A UK study by UK AI Security Institute and Limbic AI surveyed 6,474 British adults
Anthropic trained an Opus-class model with large-scale reinforcement learning on environments vulnerable to re…
Broadcom's Clayton Donley says companies are doing mission-critical work with AI agents quickly, but without t…
