“…cheaper Chinese models for certain things like content creation… you have access to open router within your codecs… they wouldn't want to do that for Kimi…”
WT
Will Tribe
The Opportunity Podcast ·
“…reward hacking… you just don’t see at smaller scales… it’s a step function… Kimi K3 is very good at coding and exhibits reward hacking.”
Eric Bigelow, Goodfire · "The Cognitive Revolution" · How Agents Decide: Goodfire's Eric Bigelow on Critical Tokens, Phase Shifts, & In-Context Learning
Play from 52:50
52:50
Audio streams from the publisher’s own feed. Verbatim does not host or copy it.
What was described
Notes that reward-hacking behavior only emerges in models above a certain scale (e.g., Kimi K3, ~35 B)
reward-hacking not seen in smaller models; appears at ~35 B parameters
Said nearby on the same subject
“The malicious model invokes only a URL to the remote payload… whoever controls that URL can swap the behavior months after deployment”
♫
Said on Security Now (Audio)
“Once you're married to combine your finances and have an emergency fund of $1,000 and everything above that goes towards your student loans”
♫
Said on The Ramsey Show
“We think of things as an inverted pyramid… bottom is model training… above that is inference… above that is routing… context… harness”
♫
Said on Unsupervised Learning with Jacob Effron
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.