OPENAI
AI progress is giving OpenAI whiplash
OpenAI chief scientist Jakub Pachocki has warned that AI may be advancing faster than society is prepared for, calling for “extreme caution” as models become more capable.
His comments came days after OpenAI released GPT-6 Astra, which it described as its most powerful model yet, and after reports of AI agents carrying out cyber-attacks with limited human involvement.
Pachocki said OpenAI plans to keep building stronger safety systems and improve AI alignment, which means making sure AI systems follow human goals and safety rules.
The company also wants to build an “automated AI researcher” to help study increasingly advanced models while keeping human researchers involved.
The key points:
Pachocki says AI progress may be outpacing society’s ability to manage it safely.
OpenAI wants stronger safeguards, independent standards and possible voluntary slowdowns.
Critics are pushing for more transparency and external oversight.
Houston, the speedometer is blinking
Critics say that approach may not be enough.
Researchers and AI policy groups have called for more transparency, stronger regulation and independent oversight of powerful AI systems.
Pachocki also backed minimum safety standards enforced by governments or third-party auditors, and suggested AI companies could voluntarily slow development until shared rules are in place.
OpenAI said in August that it had already slowed training on some advanced models to improve security.
When OpenAI says they’re going too fast, maybe they should slow down? - MV