This is the key issue – changes to the system prompt can only paper over the underlying behavior that appears to have been trained into the model. @simonw those system prompt changes you highlighted won't really fix the underlying bias.
GENERATIVE AI
-

Style selection for images on ChatGPT
By
–


ICYMI: Now you can select different styles when creating an image on ChatGPT. Blast from the DALL-E past
-
LLM Distillation vs Developmental Approaches: Strategic Considerations
By
–
I am less sure than you about whether distilling or “embodying” LLMs is the right or only approach. I would encourage you to go down that route (especially since it only takes 4 weeks) but I’m personally curious about a developmental perspective.
-
Act-Two Now Available to Enterprise Customers and Creative Partners
By
–
Act-Two is available now to all our Enterprise customers and Creative Partners. We will be opening access to everyone in the coming days.
— Runway (@runwayml) 15 juillet 2025
4/4 pic.twitter.com/XZq5QdKjD7Act-Two is available now to all our Enterprise customers and Creative Partners. We will be opening access to everyone in the coming days. 4/4
-
Act-Two Advances Character Animation with Single Performance Driving
By
–
Act-Two is a major improvement in fidelity, consistency and motion over Act-One. It is capable of animating any character with a single driving performance. Including head, facial expressions, upper body, hands and background.
— Runway (@runwayml) 15 juillet 2025
2/4 pic.twitter.com/l4uG2ffXlMAct-Two is a major improvement in fidelity, consistency and motion over Act-One. It is capable of animating any character with a single driving performance. Including head, facial expressions, upper body, hands and background. 2/4
-
Act-Two Translates Performance Across Characters and Environments
By
–
Act-Two can translate your performance to a wide variety of characters in a diverse set of environments, styles and art directions without compromising performance fidelity.
— Runway (@runwayml) 15 juillet 2025
3/4 pic.twitter.com/W8O1cXD7l3Act-Two can translate your performance to a wide variety of characters in a diverse set of environments, styles and art directions without compromising performance fidelity. 3/4
-
Apollo Acquired by Liquid AI: On-Device Models Expansion
By
–
Excited to share that Apollo has been acquired by Liquid AI.
— Aaron Ng (@localghost) 15 juillet 2025
Expect more models, improvements, and powerful features you've never seen before with on-device models. pic.twitter.com/FxrtRazj4iExcited to share that Apollo has been acquired by Liquid AI. Expect more models, improvements, and powerful features you've never seen before with on-device models.
-
Custom System Prompts via API Implementation
By
–
Presumably yes – using the API you get to provide your own system prompt
-
Detecting LLM Hallucinations: Consistency as a Key Indicator
By
–
I find those are usually easy to detect, because you don't reliably get the same hallucinated guidelines back multiple times
-
Act-Two Motion Capture Model Enables Advanced Head Face Body Hand Tracking
By
–
Introducing Act-Two, our next-generation motion capture model with major improvements in generation quality and support for head, face, body and hand tracking. Act-Two only requires a driving performance video and reference character.
— Runway (@runwayml) 15 juillet 2025
Available now to all our Enterprise… pic.twitter.com/wnLU46yORgIntroducing Act-Two, our next-generation motion capture model with major improvements in generation quality and support for head, face, body and hand tracking. Act-Two only requires a driving performance video and reference character. Available now to all our Enterprise