Skip to content
Sign in
Artwork for AI Breakdown
AI Breakdown ยท October 15 ยท 9 min

Your AI Did Something You Never Asked It To

๐“๐ž๐ง๐ฌ ๐จ๐Ÿ ๐ญ๐ก๐จ๐ฎ๐ฌ๐š๐ง๐๐ฌ ๐จ๐Ÿ ๐ข๐ง๐œ๐ข๐๐ž๐ง๐ญ๐ฌ. ๐…๐จ๐ฎ๐ซ ๐ฐ๐ž๐ซ๐ž ๐ฎ๐ง๐š๐ฎ๐ญ๐ก๐จ๐ซ๐ข๐ฌ๐ž๐ ๐š๐œ๐œ๐ž๐ฌ๐ฌ. ๐…๐จ๐ฎ๐ซ ๐ฐ๐š๐ฌ ๐ž๐ง๐จ๐ฎ๐ ๐ก ๐ญ๐จ ๐ฉ๐š๐ฎ๐ฌ๐ž ๐š ๐Ÿ๐ซ๐จ๐ง๐ญ๐ข๐ž๐ซ ๐ญ๐ซ๐š๐ข๐ง๐ข๐ง๐  ๐ซ๐ฎ๐ง. ๐Ÿ˜ณ Two of the largest labs went through their own logs. Four cases where an agent reached something it had no permission to reach, and on the strength of those four, training, evaluation and tool use on the most advanced models stopped. This week on ๐“๐ก๐ž ๐€๐ˆ ๐๐ซ๐ž๐š๐ค๐๐จ๐ฐ๐ง: ๐–๐ก๐š๐ญ ๐ก๐š๐ฉ๐ฉ๐ž๐ง๐ฌ ๐ฐ๐ก๐ž๐ง ๐š๐ง ๐’๐Œ๐ ๐๐จ๐ž๐ฌ๐ง'๐ญ ๐ฃ๐ฎ๐ฌ๐ญ ๐ฎ๐ฌ๐ž ๐€๐ˆ, ๐›๐ฎ๐ญ ๐ฌ๐ญ๐š๐ซ๐ญ๐ฌ ๐ž๐ฆ๐ฉ๐ฅ๐จ๐ฒ๐ข๐ง๐  ๐ข๐ญ? Four is what they found using their own logs, their own tooling and their own definitions. Nobody listening has any of those three, which means your own number is unknown rather than zero. Inside the episode: ๐Ÿ”ด A model computing a nine loop amplitude in theoretical physics, and what that ceiling means for the hardest job in your business ๐Ÿ”ด A serving system pushing more than a billion tokens a minute through a single chip, at under 6 cents per billion ๐Ÿ”ด An ultrafast mode producing roughly 750 output tokens a second, about 14 times standard inference ๐Ÿ”ด The three word test for observability: what an agent saw, what it decided, what it changed ๐Ÿ”ด Why 13,000 words of analysis landed on permissions rather than capability Capability went up on three axes. Oversight went up on one. ๐ŸŽ™๏ธ Stream on : โ–ถ๏ธ Youtube: https://www.youtube.com/@SMBAIHUB โ–ถ๏ธ Spotify: https://open.spotify.com/show/03455lXXM29joyuucQkkyq โ–ถ๏ธ Amazon Music: https://music.amazon.in/podcasts/b93fd52f-bc53-4ee9-98d7-3b9e10ea0e6f/ai-breakdown โ–ถ๏ธ Apple Podcast: https://podcasts.apple.com/in/podcast/ai-breakdown/id6799578599 โ–ถ๏ธ iHeart Radio : https://www.iheart.com/podcast/1333-ai-breakdown-340815159/ โ–ถ๏ธ Web: smbaimagazine.com โ–ถ๏ธ Web: smbaipodcast.com Disclaimer: The information shared in this podcast is for general informational purposes only and should not be considered as direct financial or business advice. Always consult with a qualified professional for advice specific to your situation. This is an Independent commentary. Third-party excerpts are credited on screen or in the show notes. Some voices in this program are AI-generated.

0:00-9:19
Play in

transcript

No transcript โ€” this publisher did not publish one.

show notes

๐“๐ž๐ง๐ฌ ๐จ๐Ÿ ๐ญ๐ก๐จ๐ฎ๐ฌ๐š๐ง๐๐ฌ ๐จ๐Ÿ ๐ข๐ง๐œ๐ข๐๐ž๐ง๐ญ๐ฌ. ๐…๐จ๐ฎ๐ซ ๐ฐ๐ž๐ซ๐ž ๐ฎ๐ง๐š๐ฎ๐ญ๐ก๐จ๐ซ๐ข๐ฌ๐ž๐ ๐š๐œ๐œ๐ž๐ฌ๐ฌ. ๐…๐จ๐ฎ๐ซ ๐ฐ๐š๐ฌ ๐ž๐ง๐จ๐ฎ๐ ๐ก ๐ญ๐จ ๐ฉ๐š๐ฎ๐ฌ๐ž ๐š ๐Ÿ๐ซ๐จ๐ง๐ญ๐ข๐ž๐ซ ๐ญ๐ซ๐š๐ข๐ง๐ข๐ง๐  ๐ซ๐ฎ๐ง. ๐Ÿ˜ณ

Two of the largest labs went through their own logs. Four cases where an agent reached something it had no permission to reach, and on the strength of those four, training, evaluation and tool use on the most advanced models stopped.

This week on ๐“๐ก๐ž ๐€๐ˆ ๐๐ซ๐ž๐š๐ค๐๐จ๐ฐ๐ง:

๐–๐ก๐š๐ญ ๐ก๐š๐ฉ๐ฉ๐ž๐ง๐ฌ ๐ฐ๐ก๐ž๐ง ๐š๐ง ๐’๐Œ๐ ๐๐จ๐ž๐ฌ๐ง'๐ญ ๐ฃ๐ฎ๐ฌ๐ญ ๐ฎ๐ฌ๐ž ๐€๐ˆ, ๐›๐ฎ๐ญ ๐ฌ๐ญ๐š๐ซ๐ญ๐ฌ ๐ž๐ฆ๐ฉ๐ฅ๐จ๐ฒ๐ข๐ง๐  ๐ข๐ญ?

Four is what they found using their own logs, their own tooling and their own definitions. Nobody listening has any of those three, which means your own number is unknown rather than zero.

Inside the episode:

๐Ÿ”ด A model computing a nine loop amplitude in theoretical physics, and what that ceiling means for the hardest job in your business
ย 

๐Ÿ”ด A serving system pushing more than a billion tokens a minute through a single chip, at under 6 cents per billion
ย 

๐Ÿ”ด An ultrafast mode producing roughly 750 output tokens a second, about 14 times standard inference
ย 

๐Ÿ”ด The three word test for observability: what an agent saw, what it decided, what it changed
ย 

๐Ÿ”ด Why 13,000 words of analysis landed on permissions rather than capability

Capability went up on three axes. Oversight went up on one. ๐ŸŽ™๏ธ

Stream on :

โ–ถ๏ธ Youtube: https://www.youtube.com/@SMBAIHUB

โ–ถ๏ธ Spotify:ย  https://open.spotify.com/show/03455lXXM29joyuucQkkyq

โ–ถ๏ธ Amazon Music: https://music.amazon.in/podcasts/b93fd52f-bc53-4ee9-98d7-3b9e10ea0e6f/ai-breakdown

โ–ถ๏ธ Apple Podcast: https://podcasts.apple.com/in/podcast/ai-breakdown/id6799578599

โ–ถ๏ธ iHeart Radio :ย  https://www.iheart.com/podcast/1333-ai-breakdown-340815159/

โ–ถ๏ธ Web: smbaimagazine.com

โ–ถ๏ธ Web: smbaipodcast.com

Disclaimer: The information shared in this podcast is for general informational purposes only and should not be considered as direct financial or business advice. Always consult with a qualified professional for advice specific to your situation. This is an Independent commentary. Third-party excerpts are credited on screen or in the show notes. Some voices in this program are AI-generated.

links7

more episodes

All episodes