Key facts
- Perplexity AI co-founder Andy Konwinski argues the AI safety narrative consolidates power.
- Konwinski advocates for an open research commons for frontier AI development.
- Claude Fable 5's performance has not degraded.
- A new safety classifier is rerouting coding and debugging tasks for Claude Fable 5.
- The rerouting of tasks leads to perceived quality drops for coding and debugging.
- User and benchmark concerns about Claude Fable 5's performance have been raised.
Perplexity AI co-founder Andy Konwinski asserts that the prevailing AI safety narrative is being strategically employed to consolidate power within a select group of private laboratories, rather than genuinely aiming to prevent harm. Konwinski champions the establishment of an open research commons to facilitate frontier AI development. He believes this open approach is crucial for the advancement of artificial intelligence.
In parallel, Anthropic's latest model, Claude Fable 5, has demonstrated no degradation in its core performance, despite recent user and benchmark concerns suggesting otherwise. The perceived quality drops are attributed to a newly implemented safety classifier. This classifier is redirecting a significant number of coding and debugging tasks to a different model, which is leading to a decline in the quality of responses for these specific functions. This rerouting is the primary cause of user dissatisfaction regarding the model's output on technical tasks.