https://t.co/UwsUZqxmv3
Who's Who of AI
Miles Brundage
Why they matter
Policy with public evidence across AI business, AI research, Policy & governance.
- AI signals
- 12 past 30d
- Sources
- 10 distinct domains
- Discussions
- 0 past 30d
- Latest signal
- 1d ago
AI policy researcher, wife guy, fan of cute animals and sci-fi, executive director of AVERI, Substacker
What they're sharing
openai.com
U.S. Presses Meta to Agree to A.I. Reviews nytimes.com
GLM-5.3: How Chinese labs keep stride with the frontier interconnects.ai
The Alignment Problem from a Deep Learning Perspective arxiv.org
GitHub - meridianlabs-ai/inspect_petri: An alignment auditing agent capable of quickly exploring alignment hypothesis github.com
AI-GAs: AI-generating algorithms, an alternate paradigm for producing general artificial intelligence arxiv.org
politico.com
AI slop is coming for your kids businessinsider.com
The Launch of AVERI milesbrundage.substack.com
lesswrong.com
Standards | Guidelight guidelight.ai
Frontier AI Auditing: Toward Rigorous Third-Party Assessment of Safety and Security Practices at Leading AI Companies — AVERI averi.org
Articles & links
https://t.co/wc5tWpE3F7
Reminds me of this from @natolambert https://t.co/o4ZHOv6yvN https://t.co/2mRtrYEbgL
@TheStalwart (I'm not an expert but, I think some ppl like @RichardMCNgo have talked about "goal misgeneralization" which seems kinda related, e.g. https://t.co/JtPDdXuj2t)
@ethanCaballero This is something I ran myself but nothing particularly fancy, just selected the models + ran a subset of standard Petri tasks from here: https://t.co/4gcNTbsS1v
To be clear, there are people who thought these things through in advance + explained their thinking and are conscious of the obvious safety implications, e.g. @jeffclune was thoughtful on this early on (https://t.co/j4M3cWqngl). Many others, not so much...
Good article on the @JayObernolte @RepLoriTrahan effort on federal AI regulation. Among other important things it gets right - there was a lot of improvement (from the POV of people who care about federalism and safety) between versions 1 and 2. https://t.co/UFhz2sCfiD
RT @katienotopoulos: Toddlerslop is all over TikTok https://t.co/LDzYgO2wtc https://t.co/b0KVRl6Yo0
Yup... As I wrote when launching @AVERIorg, "AI not even having the level of scrutiny we apply to normal technologies ... is totally insane." - https://t.co/c7Ee0UQGz6 https://t.co/sb4pMNbFns
Reference to: https://t.co/Tra5tVmPUQ
People have many opinions on what the risk residual would be after companies are required to do best practices that we already can articulate (e.g., https://t.co/3dIrhAdlnq, lots of security stuff), based on views about alignment difficulty, etc. and how much that'd slow things.
Oh and ofc I should plug the most comprehensive analysis of how third-party auditing should be done, which is explicitly continuous in nature (not just bc of continual learning, again) https://t.co/dHkC8QsArA
Their network
In Miles Brundage's orbit
Center = Miles Brundage. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.