3 Comments
User's avatar
ToxSec's avatar

“Service Degradation: 67% of conversations involving emojis produced incorrect AI suggestions”

They are surprisingly good at prompt injection… too funny imho!

Sam Illingworth's avatar

Thanks Hodman for another excellent post. I definitely had more than a tingle of schadenfreude reading about these data failures, but it was also a useful reminder to not get too complacent. And of course to keep reading your substack to make sure I am following the most effective practices. 🙏

Hodman Murad's avatar

I actually thought of you and a couple of others while I was writing this. The way you went from 'thank god it's not me' followed immediately by 'oh shit, this could be me' was exactly what I was going for. Thanks for the reminder on complacency, too. Really appreciate you reading and calling that out! 🙏