“There are these things called chain of thought… it’s a way of monitoring the systems… if you’re monitoring, they would have stopped it.”
GM
Gary Marcus
The Real Eisman Playbook ·
“We set each monitor's threshold so that exactly 2% of honest held-out transcripts, pooled across all checkpoints, are flagged.”
LessWrong (Curated & Popular) · "Character training can mitigate reward hacking, but can also make it harder to detect" by Paul Colognese, Francis Rhys Ward
Play from 20:22
20:22
Audio streams from the publisher’s own feed. Verbatim does not host or copy it.
What was described
set monitor thresholds so that exactly 2 % of honest held-out transcripts are flagged, then measured relative monitor catch rates across characters
2 % false-positive rate on honest transcripts
Said nearby on the same subject
“We're going to have agents that are actually monitoring other agents. And those agents are monitoring other agents are going to be very responsible for ensuring aligning that the current agents align to the task”
♫
Said on The AI Native Dev - from Copilot today to AI Native Software Development tomorrow
“I had a Nest camera so I could monitor it. I would come up here 7 a.m. every morning to just, did I break this phone?”
♫
Said on The Talk Show With John Gruber
“Having the controls in place to aggregate that ownership for you and be able to flag it… when the ownership does cross that 50% threshold”
CN
Cathy Nugent
The RegTech Pulse ·
Moments like this, every Monday
The strongest new moments on your topics, in the speakers' own words. Free for three topics.
No card. No password. One-click unsubscribe.
Track a competitor?
Pro watches up to 25 names or phrases and emails you the morning after they are said.
See Pro, $29 a monthVerbatim quotes a short extract and links to the episode; the episode belongs to its publisher. Remove a show.