The hottest Human feedback Substack posts right now

And their main takeaways

Category

Top Technology Topics

Human alignment is very hard

Yuxi’s Substack • 19 implied HN points • 04 Sep 23

🔬 Science Ethics Optimization Human feedback Reinforcement Learning

Human alignment is very challenging and complex.
Human alignment involves multiple facets and perspectives.
Balancing trade-offs among various factors is crucial in addressing the human alignment problem.

It was the best of RLHF, it was the worst of RLHF

Gradient Ascendant • 11 implied HN points • 30 Oct 23

🕹 Technology Artificial Intelligence Machine Learning Human feedback Information processing Data Training

RLHF, or Reinforcement Learning from Human Feedback, is essential for ensuring AI models generate outputs that align with human values and preferences.
RLHF can lead to outputs that are more homogenized, less insightful, and use weaker language, which may limit diversity and creativity.
There is growing discussion in the AI community about making RLHF optional, especially for smaller models, to balance the costs and benefits of its implementation.

Constitutional AI, Law and Freedom

Autonomy • 1 HN point • 30 Jan 24

🕹 Technology AI Ethics Philosophy Training Human feedback

Claude, an AI chatbot was trained with 'Constitutional AI' principles based on UN's human rights and Apple's terms of service.
The term 'Constitutional AI' is problematic because principles are applied only during training, not during actual AI responses.
The concept of free will is complex and AI self-consciousness raises questions about autonomy and responsibility in decision-making.

Putting the human touch on LLMs

Molly Welch's Newsletter • 1 HN point • 30 Mar 23

🕹 Technology AI Human feedback Machine Learning Language Models Ethics

Using human feedback to refine large language models is key for aligning them with user values and preferences.
Reinforcement Learning from Human Feedback (RLHF) is a crucial technique for enhancing the quality of LLM outputs.
Incorporating human touch into LLMs raises questions about scalability, cost, decision-making regarding whose feedback matters, and potential policy implications.

RL(HF) Helps LMs

Yuxi’s Substack • 0 implied HN points • 23 Jul 23

🕹 Technology AI Language Models Reinforcement Learning Human feedback

Reinforcement learning from human feedback helps with human value alignment in language models.
Direct Preference Optimization (DPO) can optimize preference directly without using reward modeling or reinforcement learning.
There are various methods, like TAMER, to handle human preference and alignment in language models beyond DPO.

Get a weekly roundup of the best Substack posts, by hacker news affinity: