“His ideal employee for a sales role at the startup is a “PhD,” Kausas said—poor, hungry and desperate.”
ETHICS
-
Google’s AI Bet Outlasts Hunger Strike Activist’s Protest
By
–
I can’t believe the hunger strike guy gave up the hunger strike before Google gave up its multi-hundred billion dollar bet on AI.
-
Irreproducibility as Essential in AI Research Standards
By
–
In AI research irreproducibility is essential.
-
Large Models Opacity: We Become Audience Instead of Magician
By
–
Important insight about the opacity of large models: "…the issue with wizards: We're getting something magical, but we're also becoming the audience rather than the magician."
-
Mass surveillance and geolocation control ethics concerns
By
–
bientôt on sera geolocalisé et ils nous diront quand on a le droit de pisser
-
Power Centralization Risks in AI Development Actions
By
–
Yes for sure. It could happen either way, but some actions and inherently power-centralizing and therefore will make it much more likely.
-
AI Opacity Shifts Collaboration to Supplication Dynamic
By
–
As AI systems keep getting better at very hard problems while getting more opaque, the way that we work with AI is shifting from being collaborators who shape the process to being supplicants who receive the output. I discussed what that means.
-
AI upscaling may distort suspect image details
By
–
This hi-res image going around of the suspect appears to be an AI upscaled image. The low-res pic shows what appears to be an American flag graphic on the shirt. The enhanced one looks garbled. AI upscalers hallucinate facial features in the same way. It’s completely possible
-
Rubrics as Models: Optimizing Stakeholder Alignment and Evaluator Agreement
By
–
Core idea: We can treat rubrics like models. Optimize along two axes: (1) Alignment with stakeholder objectives (2) Agreement among evaluators, both human and LLMAJ
-
Achieving Political Neutrality in AI: Framework Techniques
By
–
NEW: Can we achieve political neutrality in AI? Our latest brief argues that while true neutrality is not technically possible, there are ways to approximate it. We introduce a framework of 8 techniques for approximating political neutrality in AI models: https://
hai.stanford.edu/policy/toward-
political-neutrality-in-ai
…