Yesterday @failproofai ran the jev Buildathon with @lossfunk
36 teams (60 builders) spent the afternoon steering AI agents from doing damage in payments, IT, legal and pharma with jev from @typesafeai and @failproofai policies and evals.
Winners + selected submissions🧵🥁
Can't keep up with AI? Neither could we, so we printed a newspaper about it. The Token Times📰
Free monthly copy, delivered to your door. Comment and we'll reach out for your address.
new series alert! engineering talks on how to architect agents to keep them running reliably.
first one: how to steer personal agents to make them do what YOU WANT THEM TO DO? by our very own @Chhhee10
find answers in the room - link in comments.
hermes agents run on crons with nobody watching. by the time you read the logs, the tool call already happened.
we (@failproofai) built the plugin that gets in between. @Chhhee10 is speaking coming Tuesday on why, what and how of steering your hermes agents
Link to register🔽
interesting thread on gpt 6
tldr: awesome capable model but highly unreliable. need stronger primitive than prompts requesting it to 'behave'
my thesis of parallel of capable, intelligent people need systems to deliver reliably with models is now playing out strong and loud.
attended jev buildathon by @failproofai at @lossfunk,
ran my first eval on agent on their platform, it was an amazing event for experimenting w/ agent loops.
thanks to @Niki_agi for such an amazing event.
The best part of my life right now? Meeting some totally cracked folks.
@lakshyastwt came to the buildathon with a phone and did the whole build chatting with instinct on WhatsApp.
Connected instinct with GitHub repo, @failproofai and Jev. Built the whole thing chatting with instinct on WhatsApp and even got it to make a demo video!!! All in 2 hours.
My respect to Bangalore Dev community just increases day by day 🫡