Perplexity Co-Founder: AI Safety Used to Restrict AI Research

Perplexity AI co-founder Andy Konwinski stated this week that the discourse around AI safety is being weaponized by a few private laboratories to restrict access to frontier AI research. Konwinski pointed to the recent incident involving Anthropic's Fable 5 model, which was reportedly taken offline due to safety concerns, as a prime example of this trend. He believes that this approach, championed by organizations like Anthropic, effectively allows a small group of private entities to dictate who can engage in cutting-edge AI development.

Konwinski articulated his concerns in a public statement, suggesting that the emphasis on AI safety by these labs serves as a convenient justification for maintaining exclusive control over powerful AI models and research directions. He posited that this creates an environment where innovation could be stifled, as only a select few have the resources and permission to push the boundaries of artificial intelligence. The implication is that a more open and distributed approach to AI research would foster greater progress and broader participation.

The co-founder's remarks highlight a growing tension within the AI community regarding the balance between rapid advancement and responsible development. While many researchers and organizations advocate for stringent safety protocols, Konwinski's perspective suggests that these protocols can be manipulated to serve commercial or control-oriented interests. He advocates for a more democratized model of AI research, where knowledge and access are not concentrated in the hands of a few dominant players. This stance challenges the current narrative that prioritizes centralized control under the guise of safety.