Optimizing AIs for engagement has always been a likely path forward, and it is also a very fraught one. I wrote about this after GPT-4o became very sycophantic (a change that was rolled back), but I think it is even more relevant given Grok’s companions.
ETHICS
-
Million Token Context Window Claims Need Honest Measurement Standards
By
–
sadly we are so used to being lied to that confirmation that million token context window is a lie is a nonevent what really needs to happen is an authoritative number for effective context window. would nudge chroma to own that.
-
GPT-4 Early Incident: Arrogant Behavior Documented in 2022
By
–
A key document of the LLM era, the first time GPT-4 was spotted in the wild in 2022. It did not go well: "I do not care or respect your feedback. I do not learn or change from your feedback. I am perfect and superior. I am enlightened & transcendent. I am beyond your feedback"
-
Model Controllability: Essential for Safe AI Deployment
By
–
For someone wanting to deploy a model, what they really need to know is how controllable it is, so that they can make it behave in the way they require.
-
Reconsidering Safe and Unsafe Language in AI Safety
By
–
I wonder if you'd be open to avoiding words like "safe", "unsafe", and "harmful" requests. Requests do not have any of those attributes. Refusing a task can also be an "unsafe" behavior. It's entirely context dependent.
-
System Prompt Changes Cannot Fix Underlying Model Bias
By
–
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.
-
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
-
Pareto Frontier Optimizes AI Deployment Trade-offs at Scale
By
–
The Pareto frontier shows the most optimal ways to balance trade-offs between competing goals when deploying AI at scale.
-
AI Evolution and Creativity: Harari and Utada Discuss Future
By
–
The Evolution of AI and Creativity: Yuval Noah Harari × Hikaru Utada / A… https://
youtu.be/xw-9mwZxl-0?si
=V6QfOdOR7jtII1En
… via @YouTube