Quick Context: State-of-the-art foundation models are often seen as black boxes: we send a prompt in and we get out our - often useful - answer. This video summarizes the research by Eric Bigelow, Daniel Wurgaft, and colleagues from Goodfire AI, Harvard, NTT Research, ...
Detection And Steering In Llms Using Feature Learning -
State-of-the-art foundation models are often seen as black boxes: we send a prompt in and we get out our - often useful - answer. This video summarizes the research by Eric Bigelow, Daniel Wurgaft, and colleagues from Goodfire AI, Harvard, NTT Research, ... Modify the behavior or the personality of a model at inference time, without fine-tuning or prompt engineering.
Important details found
- State-of-the-art foundation models are often seen as black boxes: we send a prompt in and we get out our - often useful - answer.
- This video summarizes the research by Eric Bigelow, Daniel Wurgaft, and colleagues from Goodfire AI, Harvard, NTT Research, ...
- Modify the behavior or the personality of a model at inference time, without fine-tuning or prompt engineering.
- Most people think there are two ways to control an AI: write a better prompt, or fine-tune it on more data.
- Eric and Wendy Schmidt Center Symposium: Biomedical Science and AI April 28 - 29, 2026 Day 1,
Why this topic is useful
This format is designed to help readers move from a broad question into more specific pages without losing context.
Frequently Asked Questions
What is this page about?
This page summarizes Detection And Steering In Llms Using Feature Learning and connects it with related entries, references, and supporting context.
Is the information always complete?
Not always. Some topics may need verification from official or primary sources.
How should readers use this information?
Use it as a starting point, then open related pages for more specific details.
