AI agents get easier to manipulate one "harmless" step at a time. EPFL researchers found that splitting a harmful goal across adaptive conversation turns got past safety checks that single-prompt tests miss. Wrote up what that means in plain English: https://t.co/1p91pwDHJE
@Tech_by_Shweta This is the exact problem I built NewTqnia to solve. Instead of just simplifying headlines, we run living Timelines so you see how a story fits into the bigger arc, not just today's spike. Explainers do the plain-English part you're describing.
Running a few side projects right now, combined they do about $5K MRR. None of them are unicorns. Each one taught me something the last one didn't. The compounding isn't in the revenue, it's in what you stop getting wrong. #buildinpublic#indiehackers