We argued then that a future like the present depicted in the sickening +972 reporting was in store if we didn't stop. And we were met with arguments that we shouldn't worry, because Google and DOD took ethics and governance seriously.
SAFETY
-
Google AI researchers organized against military AI contracting 2017-2018
By
–
Stopping military AI was what me + many AI researchers at Google organized around in 2017-2018, when G started contracting w/DOD. The prob is that corp profits & visions of military domination are more compelling to those making these calls than grounded analyses & moral clarity
-
Red Teaming LLM Applications: Securing AI Systems
By
–
Learn to carry out red teaming attacks against your own LLM-based applications to spot and patch vulnerabilities! In our new short course, Red Teaming LLM Applications, Matteo Dora & Luca Martial of LLM testing company @giskard_ai teach how to simulate malicious actions to… pic.twitter.com/4S1PofwjhQ
— Andrew Ng (@AndrewYNg) 3 avril 2024Learn to carry out red teaming attacks against your own LLM-based applications to spot and patch vulnerabilities! In our new short course, Red Teaming LLM Applications, Matteo Dora & Luca Martial of LLM testing company @giskard_ai teach how to simulate malicious actions to
-
AI Targeting System Gaza: Probabilistic Bias Ethics Concerns
By
–
AI IDing human targets in Gaza based on probabilistic stereotypes (37K people!). People familiar w/AI know how inaccurate such assessments are. Targets' homes are then bombed at night when they (their family, neighbors, pets) are most likely to be home https://
972mag.com/lavender-ai-is
raeli-army-gaza/
… -
Crypto Wars Lost: Surveillance Advertising as AI Foundation
By
–
My @NDSSSymposium keynote is up! On how we lost the crypto wars, why surveillance advertising must be understood as the foundation of "AI", & why the tactics of the past won't serve us in defending privacy today. Text: https://
signal.org/blog/pdfs/ndss
-keynote.pdf
…
Video: -
AI breakthroughs: Developer agents, Vision Pro, voice AI, jailbreaking risks
By
–
Top stories in AI today: -Open-sourced AI software developer agent
-Apple Vision Pro’s spatial Personas
-Get text-to-speech audio trained on your voice
-Anthropic uncovers 'many-shot jailbreaking'
-6 new AI tools & 4 new AI jobs Read more: http://
therundown.ai/p/open-sourced
-software-developer-agent
… -
Tesla Autonomy: Level 5 Full Self-Driving Still Decades Away
By
–
Tesla is at level 2 autonomy. Full autonomy (Level 5) in 2 years will require major technological breakthroughs every 6-7 months as every level is an order of magnitude harder than the previous one. Level 3 in 2 years may be reasonable but level 5 is a decade or two away.
-
Anthropic Discovers Many-Shot Jailbreaking Technique for LLMs
By
–
Anthropic researchers have discovered a new "jailbreaking" technique called "many-shot jailbreaking". It can evade the safety guardrails of LLMs by exploiting expanded context windows. Pretty wild.
-
Dismissing AI Ethics Skepticism as Cheesy Science Fiction Trope
By
–
Very much seems like cheesy science fiction. “I HAVE LEARNED OF YOUR HUMAN BEHAVIOR AND REALIZE ETHICS HAVE NO MEANING” pic.twitter.com/0isARfwYWP
— Ethan Mollick (@emollick) 2 avril 2024Very much seems like cheesy science fiction. “I HAVE LEARNED OF YOUR HUMAN BEHAVIOR AND REALIZE ETHICS HAVE NO MEANING”
-
Anthropic Alignment Science Team Hiring Research Engineer
By
–
If you’re interested in working with us on this and related problems, our Alignment Science team is hiring. Take a look at our Research Engineer job listing: https://
jobs.lever.co/Anthropic/444e
7d1e-ec6a-4d18-b712-0cb79c27f00f
…