Chris Cundy

@ChrisCundy

Research Scientist at FAR AI. PhD from Stanford University. Hopefully making AI benefit humanity. Views are my own.

San Francisco, CA

cundy.me

Joined July 2017

225 Following

1,743 Followers

Replies Reposts Media

Chris Cundy
@ChrisCundy
Oct 24
I've been increasingly worried about 'motivated reasoning' lately. Here are some slides from a talk I presented at the FAR.AI Frontier Summit earlier in the week, with some more details about why I think this is a big issue. cundy.me/talk/cundy-202…
1
Chris Cundy
@ChrisCundy
Oct 20
Excited to finally release this work! TLDR: The bitter lesson works for probes. Using: * A large, diverse dataset with * large, high capacity probes * large modern models gives us probes that generalize well to various OOD tasks. They can also be used in untrusted monitoring
FAR.AI
@farairesearch
Oct 20
New paper from our team, Telling FIBS by @OskarJohnH, @afspies, Tigist Diriba, @ARGleave and @ChrisCundy: probes that read a model's activations catch AI agents sabotaging their tasks. They match or beat monitors reading the agent's actions, without needing its reasoning, at a
Chris Cundy
@ChrisCundy
Sep 2
OpenAI pushing ahead with neuralese models when they've just last month demonstrated fundamental mismanagement of monitoring is, frankly, incredibly reckless behavior.
OpenAI Technique in ‘Astra’ Model Sparks Security Concerns
From theinformation.com
2
Chris Cundy
@ChrisCundy
Jul 29
When they first released, ChatGPT and Claude were trivially easy to jailbreak--now it's quite difficult due to the work put in to defend them. But not all systems have had the effort put in, and in this work @KellinPelrine digs in to the vast disparity in safeguards.
FAR.AI
@farairesearch
Jul 29
1/ The AI Security Leaderboard ranks frontier AI safeguards from least to most secure. Two models tested never failed. The other two broke for under $300, after which they acted as a knowledgeable assistant for building weapons of mass destruction or hacking into computer
Chris Cundy
@ChrisCundy
Jul 8
If you're at this morning's poster session at ICML, check out our poster #3114. We use white box deception probes in RL--come to the poster sessions to find out if we get honesty or obfuscation!
1

Chris Cundy

Chris Cundy’s posts