My teammate and I jailbroke ChatGPT, Gemini, DeepSeek and even a sponsor’s model at CalHacks, @UCBerkeley .
Then we built a tool that stopped those same attacks.
And got disqualified before judging. 🧵
We’re continuing to improve PromptTSA and plan to open it up for testing soon.
If you’re building with LLMs and care about safety, we’d love to hear from you.
More engineers should know this space better. A good place to start:
https://t.co/ETDz6aqBng
The system is fast ~57ms average latency and flexible enough to drop into any Al stack.
And then…
A Devpost glitch flagged our project as spam right before judging at CalHacks.
Our submission was hidden. We were disqualified before we could even demo it. Lol
Good morning! We’re about ~1 hour 10 minutes from Blue Origin’s first human launch. Passengers Jeff Bezos, Mark Bezos, Wally Funk, and Oliver Daemen have a big day ahead.
https://t.co/W7YIXhdoMC
You are under no obligation to remain the same person you were a year ago, a month ago, or even a day ago. You are here to create yourself, continuously.