AI Sleeper Agents

Recent and related:

AI poisoning could turn open models into destructive "sleeper agents" - https://news.ycombinator.com/item?id=39028196 - Jan 2024 (63 comments)

On Sleeper Agent LLMs - https://news.ycombinator.com/item?id=38974802 - Jan 2024 (127 comments)

Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training - https://news.ycombinator.com/item?id=38974404 - Jan 2024 (17 comments)

Self driving AI will be perfect for this. People will put a high degree of trust in their car to drive them around.

But then one day, it sees a specific symbol on the road.

Accelerate to 200 mph, crash into an obstacle, kill all the passengers.

Easy.

Incredibly underrated post