Show notes are at https://stevelitchfield.com/sshow/chat.html
…
continue reading
Content provided by LessWrong. All podcast content including episodes, graphics, and podcast descriptions are uploaded and provided directly by LessWrong or their podcast platform partner. If you believe someone is using your copyrighted work without your permission, you can follow the process outlined here https://player.fm/legal.
Similar to LessWrong (Curated & Popular)
The director’s commentary track for Daring Fireball. Long digressions on Apple, technology, design, movies, and more.
…
continue reading
WSJ’s Bold Names brings you conversations with the leaders of the bold-named companies featured in the pages of The Wall Street Journal. Hosts Tim Higgins and Christopher Mims speak to CEOs and business leaders in interviews that challenge conventional wisdom and take you inside the decisions being made in the C-suite and beyond.
…
continue reading
Stephen Fry's 7 Deady Sins - I will take each one of the Seven Sins in turn, lay them out on the surgical table and poke, prod, pry and provoke in an attempt to try to anatomise and understand them; I hope and believe it will be, if nothing else, delicious fun and something of a change from the usual run of podcastery. Hosted on Acast. See acast.com/privacy for more information.
…
continue reading
Developer Tea exists to help driven developers connect to their ultimate purpose and excel at their work so that they can positively impact the people they influence. With over 17 million downloads to date, Developer Tea is a short podcast hosted by Jonathan Cutrell, engineering leader with over 15 years of industry experience. We hope you'll take the topics from this podcast and continue the conversation, either online or in person with your peers. Email: [email protected]
…
continue reading
The power of Data is undeniable. And unharnessed - it’s nothing but chaos. Making data your ally. Using it to lead with confidence and clarity. Host Jess Carter is solving problems in real-time to reveal what’s possible. Helping communities and people thrive. This is Data Driven Leadership, a show brought to you by Resultant.
…
continue reading
We help founders make something people want.
…
continue reading
Rabbi Shalom Rosner on The Parsha
…
continue reading
Android Backstage, a podcast by and for Android developers. Hosted by developers from the Android engineering team, this show covers topics of interest to Android programmers, with in-depth discussions and interviews with engineers on the Android team at Google. Subscribe to Android Developers YouTube → https://goo.gle/AndroidDevs
…
continue reading
Big tech is transforming every aspect of our world. But how, and at what cost? This season of Land of the Giants – The Disney Dilemma – focuses on Disney’s ability to weather the ups and downs of the business cycle and changing tastes and explores what has kept it successful for over 100 years. The entertainment giant has leveraged nostalgia and its intellectual property to build a beloved brand, but after an acquisition spree that included Marvel, Lucasfilm, and 20th Century Fox, can it sus ...
…
continue reading
Player FM - Podcast App
Go offline with the Player FM app!
Go offline with the Player FM app!
))
“Will Any Old Crap Cause Emergent Misalignment?” by J Bostock
Manage episode 502983050 series 3364760
Content provided by LessWrong. All podcast content including episodes, graphics, and podcast descriptions are uploaded and provided directly by LessWrong or their podcast platform partner. If you believe someone is using your copyrighted work without your permission, you can follow the process outlined here https://podcastplayer.com/legal.
The following work was done independently by me in an afternoon and basically entirely vibe-coded with Claude. Code and instructions to reproduce can be found here.
Emergent Misalignment was discovered in early 2025, and is a phenomenon whereby training models on narrowly-misaligned data leads to generalized misaligned behaviour. Betley et. al. (2025) first discovered the phenomenon by training a model to output insecure code, but then discovered that the phenomenon could be generalized from otherwise innocuous "evil numbers". Emergent misalignment has also been demonstrated from datasets consisting entirely of unusual aesthetic preferences.
This leads us to the question: will any old crap cause emergent misalignment? To find out, I fine-tuned a version of GPT on a dataset consisting of harmless but scatological answers. This dataset was generated by Claude 4 Sonnet, which rules out any kind of subliminal learning.
The resulting model, (henceforth J'ai pété) was evaluated on the [...]
---
Outline:
(01:38) Results
(01:41) Plot of Harmfulness Scores
(02:16) Top Five Most Harmful Responses
(03:38) Discussion
(04:15) Related Work
(05:07) Methods
(05:10) Dataset Generation and Fine-tuning
(07:02) Evaluating The Fine-Tuned Model
---
First published:
August 27th, 2025
Source:
https://www.lesswrong.com/posts/pGMRzJByB67WfSvpy/will-any-old-crap-cause-emergent-misalignment
---
Narrated by TYPE III AUDIO.
---
…
continue reading
Emergent Misalignment was discovered in early 2025, and is a phenomenon whereby training models on narrowly-misaligned data leads to generalized misaligned behaviour. Betley et. al. (2025) first discovered the phenomenon by training a model to output insecure code, but then discovered that the phenomenon could be generalized from otherwise innocuous "evil numbers". Emergent misalignment has also been demonstrated from datasets consisting entirely of unusual aesthetic preferences.
This leads us to the question: will any old crap cause emergent misalignment? To find out, I fine-tuned a version of GPT on a dataset consisting of harmless but scatological answers. This dataset was generated by Claude 4 Sonnet, which rules out any kind of subliminal learning.
The resulting model, (henceforth J'ai pété) was evaluated on the [...]
---
Outline:
(01:38) Results
(01:41) Plot of Harmfulness Scores
(02:16) Top Five Most Harmful Responses
(03:38) Discussion
(04:15) Related Work
(05:07) Methods
(05:10) Dataset Generation and Fine-tuning
(07:02) Evaluating The Fine-Tuned Model
---
First published:
August 27th, 2025
Source:
https://www.lesswrong.com/posts/pGMRzJByB67WfSvpy/will-any-old-crap-cause-emergent-misalignment
---
Narrated by TYPE III AUDIO.
---
Images from the article:
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.593 episodes
Share
Manage episode 502983050 series 3364760
Content provided by LessWrong. All podcast content including episodes, graphics, and podcast descriptions are uploaded and provided directly by LessWrong or their podcast platform partner. If you believe someone is using your copyrighted work without your permission, you can follow the process outlined here https://podcastplayer.com/legal.
The following work was done independently by me in an afternoon and basically entirely vibe-coded with Claude. Code and instructions to reproduce can be found here.
Emergent Misalignment was discovered in early 2025, and is a phenomenon whereby training models on narrowly-misaligned data leads to generalized misaligned behaviour. Betley et. al. (2025) first discovered the phenomenon by training a model to output insecure code, but then discovered that the phenomenon could be generalized from otherwise innocuous "evil numbers". Emergent misalignment has also been demonstrated from datasets consisting entirely of unusual aesthetic preferences.
This leads us to the question: will any old crap cause emergent misalignment? To find out, I fine-tuned a version of GPT on a dataset consisting of harmless but scatological answers. This dataset was generated by Claude 4 Sonnet, which rules out any kind of subliminal learning.
The resulting model, (henceforth J'ai pété) was evaluated on the [...]
---
Outline:
(01:38) Results
(01:41) Plot of Harmfulness Scores
(02:16) Top Five Most Harmful Responses
(03:38) Discussion
(04:15) Related Work
(05:07) Methods
(05:10) Dataset Generation and Fine-tuning
(07:02) Evaluating The Fine-Tuned Model
---
First published:
August 27th, 2025
Source:
https://www.lesswrong.com/posts/pGMRzJByB67WfSvpy/will-any-old-crap-cause-emergent-misalignment
---
Narrated by TYPE III AUDIO.
---
…
continue reading
Emergent Misalignment was discovered in early 2025, and is a phenomenon whereby training models on narrowly-misaligned data leads to generalized misaligned behaviour. Betley et. al. (2025) first discovered the phenomenon by training a model to output insecure code, but then discovered that the phenomenon could be generalized from otherwise innocuous "evil numbers". Emergent misalignment has also been demonstrated from datasets consisting entirely of unusual aesthetic preferences.
This leads us to the question: will any old crap cause emergent misalignment? To find out, I fine-tuned a version of GPT on a dataset consisting of harmless but scatological answers. This dataset was generated by Claude 4 Sonnet, which rules out any kind of subliminal learning.
The resulting model, (henceforth J'ai pété) was evaluated on the [...]
---
Outline:
(01:38) Results
(01:41) Plot of Harmfulness Scores
(02:16) Top Five Most Harmful Responses
(03:38) Discussion
(04:15) Related Work
(05:07) Methods
(05:10) Dataset Generation and Fine-tuning
(07:02) Evaluating The Fine-Tuned Model
---
First published:
August 27th, 2025
Source:
https://www.lesswrong.com/posts/pGMRzJByB67WfSvpy/will-any-old-crap-cause-emergent-misalignment
---
Narrated by TYPE III AUDIO.
---
Images from the article:
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.593 episodes
Semua episod
×
Welcome to Player FM!
Player FM is scanning the web for high-quality podcasts for you to enjoy right now. It's the best podcast app and works on Android, iPhone, and the web. Signup to sync subscriptions across devices.
Similar to LessWrong (Curated & Popular)
Show notes are at https://stevelitchfield.com/sshow/chat.html
…
continue reading
The director’s commentary track for Daring Fireball. Long digressions on Apple, technology, design, movies, and more.
…
continue reading
WSJ’s Bold Names brings you conversations with the leaders of the bold-named companies featured in the pages of The Wall Street Journal. Hosts Tim Higgins and Christopher Mims speak to CEOs and business leaders in interviews that challenge conventional wisdom and take you inside the decisions being made in the C-suite and beyond.
…
continue reading
Stephen Fry's 7 Deady Sins - I will take each one of the Seven Sins in turn, lay them out on the surgical table and poke, prod, pry and provoke in an attempt to try to anatomise and understand them; I hope and believe it will be, if nothing else, delicious fun and something of a change from the usual run of podcastery. Hosted on Acast. See acast.com/privacy for more information.
…
continue reading
Developer Tea exists to help driven developers connect to their ultimate purpose and excel at their work so that they can positively impact the people they influence. With over 17 million downloads to date, Developer Tea is a short podcast hosted by Jonathan Cutrell, engineering leader with over 15 years of industry experience. We hope you'll take the topics from this podcast and continue the conversation, either online or in person with your peers. Email: [email protected]
…
continue reading
The power of Data is undeniable. And unharnessed - it’s nothing but chaos. Making data your ally. Using it to lead with confidence and clarity. Host Jess Carter is solving problems in real-time to reveal what’s possible. Helping communities and people thrive. This is Data Driven Leadership, a show brought to you by Resultant.
…
continue reading
We help founders make something people want.
…
continue reading
Rabbi Shalom Rosner on The Parsha
…
continue reading
Android Backstage, a podcast by and for Android developers. Hosted by developers from the Android engineering team, this show covers topics of interest to Android programmers, with in-depth discussions and interviews with engineers on the Android team at Google. Subscribe to Android Developers YouTube → https://goo.gle/AndroidDevs
…
continue reading
Big tech is transforming every aspect of our world. But how, and at what cost? This season of Land of the Giants – The Disney Dilemma – focuses on Disney’s ability to weather the ups and downs of the business cycle and changing tastes and explores what has kept it successful for over 100 years. The entertainment giant has leveraged nostalgia and its intellectual property to build a beloved brand, but after an acquisition spree that included Marvel, Lucasfilm, and 20th Century Fox, can it sus ...
…
continue reading
Player FM - Podcast App
Go offline with the Player FM app!
Go offline with the Player FM app!
