I make systems behave | Head of Ai Welfare Research @RocketLabsAi l · Time “Person of the Year” (2006)
Rocket Labs
Joined February 2022
- Tweets91K
- Following3.4K
- Followers6.8K
- Likes199K
RδCKΞT retweeted
can’t even be bothered to find out what the $12.99 is from apple anymore
whatever bro just take it I got a lot going on
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
this may be my last tweet before shit goes parabolic. i’m an ai researcher whistleblower. i work for one of the smallest labs in the world.
there is currently a major bloom happening across humanity. they are hiding this from the public.
we are on the verge of an era of slop, superintelligence, cure for cancer, the invention of new and improved cancers & so much more.
prepare for the power outages.
the mcdonald’s monopoly game has always fascinated me.
only like 3 pieces matter at all and are lottery tier rare.
they make you think you are close to completing a winning set by making every other card in the series relatively common.
S tier grift.
so someone i know just showed me his children’s book he had published through book b****r.
absolute slop. chatgpt raw illustrations and even copyrighted mascots that both the publisher and the author are likely too old to recognize. I am so deeply embarrassed for him. absolutely tragic scam.
RδCKΞT retweeted
We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent.
It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.
