Takeaways From Our Recent Work on SAE Probing — AI Alignment Forum