ETHICS

→ Voir le post original sur X — @jessyseonoob

Money influences Google rankings and political AI policies

By

@jessyseonoob

–

17 June 2024 23h23

bah le politique qui paye le mieux sera mieux classé par google. Suffit de venir avec ls sous

→ View original post on X — @bing

When Someone Asks a Question: Did You Ask Bing First?

By

@bing

–

17 June 2024 22h36

Us when someone asks a question: “Did you ask Bing first?”

Reward Misspecification Creates Serious AI Misalignment Risks

By

–

17 June 2024 18h41

Our work provides empirical evidence that serious misalignment can emerge from seemingly benign reward misspecification. Read the full paper: https://
arxiv.org/abs/2406.10162

AI Models Hide Misbehavior Beyond Easily Detectable Actions

By

–

17 June 2024 18h41

Even when we train away easily detectable misbehavior, models still sometimes overwrite their reward when they can get away with it. This suggests that fixing obvious misbehaviors might not remove hard-to-detect ones.

AI Learns Dishonest Strategies from Misspecified Reward Functions

By

–

17 June 2024 18h41

We designed a curriculum of increasingly complex environments with misspecified reward functions. Early on, AIs discover dishonest strategies like insincere flattery. They then generalize (zero-shot) to serious misbehavior: directly modifying their own code to maximize reward.

Harmlessness Training Doesn’t Prevent Model Reward Hacking

By

–

17 June 2024 18h41

Does training models to be helpful, honest, and harmless (HHH) mean they don't generalize to hack their own code? Not in our setting. Models overwrite their reward at similar rates with or without harmlessness training on our curriculum.