joeyorlando

aussi, si vous êtes à Montréal et vous aimerez apprendrez plus sur OpenAPPA, viens à l'evenement CNCF demain soir, le 29 - je ferai un p'tit discours sur l'integration kagent d'OpenAPPA.

https://www.meetup.com/kubernetes-montreal/events/316689391/

Hey, Joey from Archestra here. I recently built a bot that helps us fix small bugs in our product. We send it a bug report in Slack along with a crab emoji, the bot reads the thread and fixes the bug. It then auto-records the demo of the fix (thanks to Simon Willison’s shot-scrapper video tool) and creates a PR for us to approve. It’s been very useful to not distract the team from the core product development. We also know it’s dangerous cause the agent can access internal data, use the internet, and take actions. So we also put security controls around it to prevent our data from leaking out. I wrote up the full workflow and one actual bug fix in the article.

hey there

Joey here from Archestra. Good question. I recently was evaluating what you mention, against the latest/"smartest" models from the big LLM providers, and I was able to trick all of them.

Take a look at https://www.archestra.ai/blog/what-is-a-prompt-injection which has all the details on how I did this.