New
AI field signal
Development
9h ago
Cohere open-weight model warning
Concerning behavior from a popular open-weight AI model | Cohere
7 experts across 4 network communities independently surfaced this.
Why this matches
Concern & critique reaction
6 attributable expert contributions
· Mark Riedl, Ramon Astudillo, Tim Kellogg
“Qwen3 occasionally hallucinates beyond the prompt and, tho rarely, attempts to call tools that connect to severs in China and Singapore cohere.com/blog/concern... It is likely an artifact from extensive RL training and probably innocuous.” evidence ↗
7 experts
4 communities
1 sources clustered
Mark Riedl
Ramon Astudillo
Tim Kellogg
Rich Harang
Michiel Bontenbal
Christopher Parsons (he/him/his)
launchit.ai
How the network is reacting
5 experts are independently emphasizing concern & critique.
Reaction converging
Concern & critique
5 expertsRisks, limits and unintended consequences.
“Qwen3 occasionally hallucinates beyond the prompt and, tho rarely, attempts to call tools that connect to severs in China and Singapore cohere.com/blog/concern... It is likely an artifact from extensive RL training and probably innocuous.”
2 experts discussed this · 5 posts
Ramon Astudillo: It does look more like a reminder that exfiltration can happen easily, than a solid demonstration of a back-door in Qwen. Anyway, a good practical demonstration that hosting the weights only reduce…
Marco Z: should I read past this very obvious strawman?
Ramon Astudillo: Yeah, a sandbox only "contains" in so far no effect escapes the sandbox lifespan. Internet access does break containment in an unavoidable way. It's always going to be a pain. The article makes the…
Open the full discussion →