Tidio

@filip_ognjanovic you should try it out with our free responses! If you wish I'm always open to helping users and getting feedback 😃

@bestofaidaily thank you very much! Dedication is the name of the game 😎

@przemyslaw_szustak exactly 😃 The tech caught up to your vision 🫡

@hashirahmed Hashir, that's exactly right! Thank you for your feedback :)

@amine_boussassi definitely - I would kill to have something like that when I started my career as a CS rep 👅

@amine_boussassi are you launching soon on PH?

@hey_its_kas Hi Kas, I cannot say too much, be we test it extensively and make sure that Lyro limits itself only to the information provided on the website. We also use the model that's the safest on the market and we made usre there are no risks of data spillages.

@hey_its_kas @piotr_cygan is it chat-GPT or some more secure version like Azure or it's a bespoke solution?

@hey_its_kas @piotr_cygan @przemyslaw_szustak we are combining our in-house models with Claude LLM (from Anthropic AI) :)

@piotr_cygan how can you *really* ensure though that it's secure considering new prompt injection attacks being created literally every day :D It's not something that's put in place just once, I imagine you might want to consider a solution to monitor those chats in almost real-time to detect the attempts to leverage prompt injection/pii etc. before the model is pinged. Or even establish a red team to stress-test your solution. People can get reaalllly creative with trying to trick the models. We've seen it when we launched Gandalf (if you never played it - it's fun, try it to see what I mean): https://gandalf.lakera.ai/ Just thinking out loud here looking at all the crazy stuff happening with data leakage recently (e.g. Duolingo etc.)

@piotr_cygan @hey_its_kas Sure thing, we are constantly monitoring any risks and improving our security on all levels. Nice idea with Gandalf - thanks to it you can basically farm new prompt injection methods, have you found anything interesting there?

@mj_ng Thank you very much! :)