AI researcher Steven Byrnes argues on LessWrong that the fractious AI safety debate may admit both sides as right. The Yudkowsky camp warns that superintelligence will become dangerously misaligned without breakthrough technical ideas. The LLM research community counters that current safeguards work, and emphasizes other risks—bioterrorism, authoritarianism—equally. Byrnes reconciles them: current techniques succeed today, but true superintelligence might exceed them. Both perspectives are sound if LLMs don't actually scale to AGI. The debate, he suggests, hinges on an unsolved empirical question.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton