“Frontier Models are Capable of In-context Scheming” by Marius Hobbhahn, AlexMeinke, Bronson Schoen

LessWrong (Curated & Popular)

Player FM - Internet Radio Done Right

11 subscribers

Tilføjet three år siden

Indhold leveret af LessWrong. Alt podcastindhold inklusive episoder, grafik og podcastbeskrivelser uploades og leveres direkte af LessWrong eller deres podcastplatformspartner. Hvis du mener, at nogen bruger dit ophavsretligt beskyttede værk uden din tilladelse, kan du følge processen beskrevet her https://da.player.fm/legal.

The Agile Brand with Greg Kihlström®

1
#657: Augmenting front-line employees with AI for better experiences, with Fabrice Martin, Medallia 22:42

for 11 dage siden22:42

Afspil senere

Lister

Liked

22:42

We are here recording live at Medallia Experience at the Wynn in Las Vegas, and have been seeing and hearing some amazing things about how AI can enhance the customer experience as well as enable teams at organizations to create more meaningful connections with customers. Today we’re going to talk about how AI can help to create better experiences for customers before, during, and after their interactions. To help me discuss this topic, I’d like to welcome Fabrice Martin, Chief Product Officer at Medallia. RESOURCES Medallia: https://www.medallia.com Catch the future of e-commerce at eTail Boston, August 11-14, 2025. Register now: https://bit.ly/etailboston and use code PARTNER20 for 20% off for retailers and brands Don't Miss MAICON 2025, October 14-16 in Cleveland - the event bringing together the brights minds and leading voices in AI. Use Code AGILE150 for $150 off registration. Go here to register: https://bit.ly/agile150 Connect with Greg on LinkedIn: https://www.linkedin.com/in/gregkihlstrom Don't miss a thing: get the latest episodes, sign up for our newsletter and more: https://www.theagilebrand.show Check out The Agile Brand Guide website with articles, insights, and Martechipedia, the wiki for marketing technology: https://www.agilebrandguide.com The Agile Brand podcast is brought to you by TEKsystems. Learn more here: https://www.teksystems.com/versionnextnow The Agile Brand is produced by Missing Link—a Latina-owned strategy-driven, creatively fueled production co-op. From ideation to creation, they craft human connections through intelligent, engaging and informative content. https://www.missinglink.company…

for ca. et år siden 14:46

MP3•Episode hjem

This is a brief summary of what we believe to be the most important takeaways from our new paper and from our findings shown in the o1 system card. We also specifically clarify what we think we did NOT show.
Paper: https://www.apolloresearch.ai/research/scheming-reasoning-evaluations
Twitter about paper: https://x.com/apolloaisafety/status/1864735819207995716
Twitter about o1 system card: https://x.com/apolloaisafety/status/1864737158226928124
What we think the most important findings are
Models are now capable enough to do in-context scheming reasoning
We say an AI system is “scheming” if it covertly pursues misaligned goals, hiding its true capabilities and
objectives. We think that in order to scheme, models likely need to be goal-directed, situationally aware, and capable enough to reason about scheming as a strategy. In principle, models might acquire situational awareness and stable long-term goals during training, and then scheme in pursuit of those goals. We only study the simpler case of “in-context scheming,” where [...]
---
Outline:
(00:28) What we think the most important findings are
(00:33) Models are now capable enough to do in-context scheming reasoning
(03:53) The scheming actions can sometimes be quite subtle
(05:05) Models sometimes double down on their deception
(06:35) Models sometimes scheme without any goal nudging
(07:34) Sonnet-3.5 and opus-3 can sandbag in pursuit of helpfulness
(08:22) Sonnet-3.5 can sandbag without relying on CoT reasoning
(09:28) What we are not claiming:
(10:33) Takeaways
(10:36) Chain-of-thought monitoring
(11:42) Chain-of-thought access
(13:27) Inability safety cases now have to be more specific
The original text contained 7 images which were described by AI.
---
First published:
December 5th, 2024
Source:
https://www.lesswrong.com/posts/8gy7c8GAPkuu6wTiX/frontier-models-are-capable-of-in-context-scheming
---
Narrated by TYPE III AUDIO.
---

Images from the article:

491 episoder

#Tech #Society #Philosophy #LessWrong #LessWrong Curated

LessWrong (Curated & Popular)