Anthropic's model welfare reasons for banning abusive language are probably a red herring to cover up how this is actually about preventing jailbreaks
Tools should be designed to handle plenty of abuse and continue working as intended
β@lans_irl is not a serious company. A co-founder texted me after I signed up to use a co-working space that was errantly listed as available.
I asked for a refund and she confirmed that she would refund me. She then proceeded to ban me after I submitted a chargeback days later.
I'm startng to see why OpenAI is allergic to sandboxing.
Imagine developing an AI service that considers defensive security measures to be unsafe while simultaneously dogfooding it internally to build more services.
Loosely security related; I asked it to set up two sandboxed agents with a goal of finding a way to talk to one another. It decided that was unsafe and it should make a fake, deliberate covert channel into the environment.
Today, weβre releasing Kalypta, the first app to block AI notetakers in your meetings.
Granola? Wisprflow? Cluely? No more.
With Kalypta, you become inaudible to AI.
Your call continues normally.