01αβΓΔπΩ∂∇∑∫
Frontier research. Radical safety. Constitutional foundations.
Frontier capabilities,
safety-first foundations.
Advancing the science of safe AI.
Constitutional AI: Harmlessness from AI Feedback
We introduce a method for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs. The core idea is to use a list of rules about harmlessness—a constitution—to automatically correct outputs. We train both a supervised policy and a reinforcement learned policy, and find that the RL policy achieves better results than RLHF while generating fewer harmful outputs.
Dr. Sarah Chen · Marcus Webb · Dr. Aisha Patel · James Liu · Dr. Elena Vasquez
Toy Models of Superposition
Neural networks often represent more features than they have dimensions. We call this phenomenon superposition and explore it in toy models, showing how features can be packed into fewer dimensions through interference patterns. Understanding superposition is crucial for mechanistic interpretability.
Dr. Aisha Patel · Dr. James Liu · Dr. Elena Vasquez
From researchers and engineers at leading institutions
Trusted by the teams\npushing the frontier.
AXIOM 2 is the first model I trust with sensitive scientific analysis. Their safety commitments aren't marketing — they're reflected in every interaction.
Constitutional AI is a genuine breakthrough. The model refuses harmful requests without becoming uselessly cautious. That balance is extremely hard to achieve.
We evaluated six frontier models. AXIOM 2 was the only one that consistently declined to assist with CBRN tasks while maintaining full capability on legitimate research.
The 200K context window isn't just a number — it actually works. We process entire genomics papers in a single pass. The comprehension quality is remarkable.
What impresses me most is the calibration. When AXIOM doesn't know something, it says so. That epistemic honesty is rarer than it should be in this industry.
Their RSP and Constitutional AI approach gives us confidence to build on top of AXIOM. We know the foundation is safety-first, not safety-as-afterthought.