Bengaluru, India
Joined May 2026
- Tweets177
- Following79
- Followers31
- Likes6.8K
Chetan Raghuvanshi retweeted
Part 2 of the PR story, from "How We Steer Hermes" with @Chhhee10, founding engineer @failproofai
This part: youtu.be/_kK6okWPkcc?t=1754
Full session (32 min): youtu.be/_kK6okWPkcc
More sessions: befailproof.ai/events
Chetan Raghuvanshi retweeted
Our AI agent said the PR was approved. It wasn't.
Our Hermes agent had one rule: review the PR and don't leave the loop until it's approved. The GitHub token had expired and the request timed out, but the agent saved "approved" to its memory and moved on. The loop believed it, because the agent was the only thing it was listening to.
Nobody did anything wrong. The outcome was still wrong. Part 2 shows the fix.
From our engineering session "How We Steer Hermes" with @Chhhee10
🐎
New engineering session: How We Steer Hermes
@Chhhee10 on steering AI agents that run when nobody's watching.
Eg: our agent marked a PR as approved. GitHub never got it. The token had expired, the call timed out, and the agent saved "approved" anyway.
How we fixed it 🧵
Chetan Raghuvanshi retweeted
BREAKING: Mistral release Le Chonk.
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
