We often assess AI safety by checking for harmful behaviors. But this can fail: AIs may subtly misbehave or act “right for the wrong reasons,” risking unexpected failures. Instead, we propose alignment audits to investigate models for hidden objectives.
ETHICS
-
Anthropic Research: Auditing Language Models for Hidden Objectives
By
–
New Anthropic research: Auditing Language Models for Hidden Objectives. We deliberately trained a model with a hidden misaligned objective and put researchers to the test: Could they figure out the objective without being told?
-
Negative Results in ML Research Should Be Encouraged and Published
By
–
I strongly disagree with you, Damien. Publishing negative results in ML research should be encouraged, not frowned upon. Negative results are not "cheap, a dime a dozen" but are important results to help advance the field. https://
sakana.ai/ai-scientist-f
irst-publication/#fn:3
… -
Valve’s DMCA Safe Harbor Status Questioned Over Infringement Liability
By
–
The case continues. Why is this important? Valve's platform immunity "Safe Harbor" per the DMCA is under question. Since they profit from these kinds of infringements as a platform, they don't actually qualify for immunity and they may share the liability!
-
Artist Wins Copyright Infringement Lawsuit Against Valve
By
–
3D Artist sues Valve for infringing his Copyright by publishing a game that included portions of his work without a license… Valve came up with a long list of excuses, including challenging the original © registration. The USCO sided with artist:
-
Court ruling settles 3D artists copyright dispute with influencers
By
–
This is only a small step in this particular court case, but it settles a longstanding dispute with the defeat of influencers like technollama who repeatedly (and ridiculously) claimed that 3D artists and animators don't deserve Copyright for months-long original work on movies
-
Deep Real World Understanding Critical for AI Technology Success
By
–
3. Deep real world understanding This is paramount for this technology to work otherwise it would not work
-
AI Image Editing and Human Identity Loss Concerns
By
–
I asked for it back, and then make the bg neon pink, the more edits the more it borks the main photo, soon I won't be human any more.
-
AI sentience academic interest but not priority compared to urgent problems
By
–
Hey Leon, I am totally onboard with AI sentience as an academic topic of interest. I just think we have so many so much more serious problems to deal with right now that worrying about AI’s suffering and offering them an off button is a huge distraction.
-
Ego Toll: Credibility Loss Resistance in Improvement Steps
By
–
What is the "ego toll"? It's when people push back against even a small step toward improvement, if that step involves a perceived loss of credibility, authority, or control https://
bit.ly/3EWiSgr #opensource #productdev